Using Pavlovian Higher-Order Conditioning
Paradigms to Investigate the Neural Substrates
of Emotional Learning and Memory
Jonathan C. Gewirtz1,3and Michael Davis2
1University of Minnesota, Minneapolis,Minnesota55455,USA;2Emory University, Atlanta,Georgia30322,USA
In first-order Pavlovian conditioning, learning is acquired by pairing a conditioned stimulus (CS) with an
intrinsically motivating unconditioned stimulus (US; e.g., food or shock). In higher-order Pavlovian
conditioning (sensory preconditioning and second-order conditioning), the CS is paired with a stimulus that
has motivational value that is acquired rather than intrinsic. This review describes some of the ways
higher-order conditioning paradigms can be used to elucidate substrates of learning and memory, primarily
focusing onfearconditioning.First-order conditioning,
preconditioning allow for the controlled demonstration of three distinct forms of memory, the neural
substrates of which can thus be analyzed. Higher-order conditioning phenomena allow one to distinguish
more precisely between processes involved in transmission of sensory or motor information and processes
involved in the plasticity underlying learning. Finally, higher-order conditioning paradigms may also allow one
to distinguish between processes involved in behavioral expression of memory retrieval versus processes
involved in memory retrieval itself.
By reducing learning to its most rudimentary components,
the influence of undefined and uncontrollable confounding
variables can be minimized. Consequently, much of the
progress that has been achieved in searching for the neural
substrates of learning and memory has been made using
some of the simplest forms of learning. In one such para-
digm, first-order Pavlovian conditioning, a conditioned
stimulus (CS, such as a tone or light) acquires motivational
significance by being paired with an intrinsically aversive or
rewarding unconditioned stimulus (US, such as foot shock
or food). Learning is evaluated by the ability of the CS to
elicit a conditioned response (CR) in anticipation of the
occurrence of the US. The use of first-order conditioning
has revealed genetic and cellular mechanisms underlying
learning and memory in species ranging from the fruit fly
and sea snail to the mouse and rat.
Thus far, less attention has been paid by neurobiolo-
gists to the potential uses of higher-order Pavlovian condi-
tioning, learning phenomena in which a CS (S2) acquires
associative strength by being paired with another CS (S1)
rather than with a US. The pairing of S2 with S1 may occur
before S1 is paired with the US (sensory preconditioning) or
after S1 has been paired with the US (second-order condi-
tioning; see Table 1). The cardinal feature of both sensory
preconditioning and second-order conditioning—and that
which recommends these paradigms to the service of neu-
robiologists interested in learning and memory—is that S2
acquires associative strength even though it is never paired
directly with a US.
This article will describe two promising avenues of
research in higher-order conditioning. First, the fact that
reinforcing value is acquired makes higher-order condition-
ing well suited to investigating the neural substrates of
different forms of reinforcement. Second, the absence of
direct pairings between the CS and US allows one to char-
acterize the roles played by molecular, genetic, pharmaco-
logical, and anatomical mechanisms in different stages of
learning or memory more precisely than could be achieved
using other conditioning paradigms.
Higher-Order Conditioning: Fact or Artifact?
There have been countless demonstrations of second-order
conditioning since Pavlov (1927) first showed that saliva-
tion could be conditioned to a black square by pairing it
with the sound of a metronome that had previously been
paired with food. However, to demonstrate higher-order
conditioning conclusively, two criteria should be met: the
ability of S2 to elicit a conditioned response must result
directly from the pairing of S2 with S1, and the ability of S1
to support conditioning must result directly from its pairing
(previous or subsequent) with the US (Rizley and Rescorla
1972). A large number of experiments conducted since the
1960s that have included control groups appropriate to
meet the first criterion (e.g., Kamil 1969; McAllister and
E-MAIL email@example.com; FAX (612) 626-2079.
LEARNING & MEMORY 7:257–266 © 2000 by Cold Spring Harbor Laboratory Press ISSN1072-0502/00 $5.00
McAllister 1964; Prewitt 1967; Tait et al. 1969) and several
that have included control groups appropriate to meet the
second (e.g., Pfautz et al. 1978; Rizley and Rescorla 1972)
have still revealed evidence of robust second-order condi-
tioning and sensory preconditioning. However, clear dem-
onstrations of higher-order conditioning in some studies do
not authenticate demonstrations of higher-order condition-
ing in other studies conducted under very different circum-
stances. Thus, it is important when using these procedures
to include, at least initially, control groups where S2 and S1
are explicitly unpaired in time to rule out stimulus gener-
alization (i.e., the tendency for a response to generalize
from one stimulus to another), the most likely source of
experimental artifact in higher-order conditioning para-
Although second-order conditioning and sensory pre-
conditioning are intrinsically weaker than first-order condi-
tioning they can, at least, be enhanced by ensuring that
first-order conditioning is particularly strong. This can be
achieved by using a US of high value or intensity (Helmstet-
ter and Fanselow 1989; Rescorla and Furrow 1977). With
regard to second-order conditioning, “refresher” first-order
conditioning trials can be interspersed during the phase of
second-order conditioning. This prevents the extinction of
first-order conditioning that would otherwise result from
the presentation of S1 in the absence of reinforcement.
Other factors identified as being important in determin-
ing the strength of higher-order conditioning are the simi-
larity of S1 and S2 and the temporal and physical distance
between them. In cases where the conditioned stimuli are
localized in space (e.g., differently colored key lights), sec-
ond-order conditioning is enhanced when S2 and S1 are in
close spatial proximity (Rescorla 1979; Rescorla and Cun-
ningham 1979). Second-order conditioning also can be fa-
vored when S1 and S2 are stimuli within the same sensory
modality (Rescorla and Furrow 1977; but see Holland
1977). Sensory preconditioning is most strongly acquired
when S1 and S2 are presented simultaneously, although the
superiority of this arrangement can only be detected using
a sophisticated set of control procedures (Rescorla 1980b).
However, even with the above conditions in place, one
difficulty with the use of second-order conditioning is that
learning appears typically to be transient. That is, second-
order conditioning usually reaches asymptote after a small
number of trials and begins to decline with further training.
The most plausible interpretation of this transience is that
with greater training, second-order learning is obscured by
the development of conditioned inhibition (e.g., Herendeen
and Anderson 1968; Rescorla 1973; Holland and Rescorla
1975b; Yin et al. 1994). A conditioned inhibitor signals the
nonoccurrence of reinforcement and, thus, inhibits the
elicitation of conditioned responses by excitatory stimuli. It
is not hard to see why procedures described above that
produce second-order conditioning—S1 is reinforced on ev-
ery occasion it is presented (S1-US trials) except when it is
accompanied by S2 (S2-S1-no US trials)—would also pro-
duce conditioned inhibition. In fact, this training procedure
(also called a feature-negative discrimination) is often the
procedure of choice for generating conditioned inhibition.
Hence, the transience of second-order conditioning can be
explained if one assumes that second-order conditioning
develops more rapidly than conditioned inhibition, the two
effects are antagonistic at the level of the behavioral re-
sponse, and conditioned inhibition is ultimately the stron-
ger of the two phenomena so that it obscures second-order
conditioning as the amount of training is increased (Fig. 1).
The interplay between second-order conditioning and
conditioned inhibition thus creates an inverted U-shaped
function, relating the strength of second-order conditioning
to the number of training trials. Potentially, this can lead to
interpretive problems. For example, a lesion, drug, or ge-
netic manipulation that decreased second-order condition-
ing might do so by enhancing conditioned inhibition rather
than actually impairing second-order conditioning. Para-
doxically, therefore, a treatment that actually enhanced
overall learning might appear to impair learning if one
looked only at second-order conditioning. Thus, for second-
Second-Order Conditioning, and Sensory
Phases of Training in First-Order Conditioning,
. . .
second-order conditioning, frequently observed in the literature. In
such cases, it is assumed that the size of the conditioned response
is constrained by the sum of second-order conditioning and con-
ditioned inhibition. It should be noted that negative conditioned
responses are not observed in most paradigms.
An explanation in schematic terms of the transience of
Gewirtz and Davis
order conditioning to be useful, one has to control for the
potential confound of conditioned inhibition.
One solution is to arrange training parameters that fa-
vor the development of second-order conditioning over
conditioned inhibition. Pavlov’s (1927) recommendation of
using trace conditioning has not been proven to solve this
problem (e.g., Kehoe et al. 1981). A more promising ap-
proach is the use of a partial reinforcement procedure dur-
ing first-order conditioning (i.e., a mixture of S1+ and S1−
trials, rather than S1+ trials only). This clearly has a delete-
rious impact on conditioned inhibition (Rescorla 1999),
while producing robust second-order conditioning (Kamil
1969; Gewirtz and Davis 1997a).
A second solution is to find a measure of second-order
conditioning that is not affected by the presence of condi-
tioned inhibition. Holland and Rescorla (1975b) observed
both conditioned inhibition and second-order conditioning
to the same CS in an appetitive conditioning paradigm rela-
tive to an unpaired control group. After substantial training,
S2 still elicited conditioned responding itself (i.e., second-
order conditioning), but suppressed conditioned respond-
ing to S1 when the two stimuli were presented in com-
pound (i.e., S2 also passed the summation test of condi-
tioned inhibition.) We have made a similar observation in
fear conditioning, using the fear-potentiated startle para-
digm. In this paradigm, fear is measured as the enhance-
ment in the amplitude of the startle reflex in the presence
versus the absence of a CS. Thus, the time course of the fear
response can be measured by presenting the startle-eliciting
stimulus at different time points relative to the onset and
offset of the CS (Davis et al. 1989). Using a serial S2-S1
training procedure, Falls and Davis (1997) found evidence
of second-order fear-potentiated startle when the startle re-
flex was elicited in test during S2 but of conditioned inhi-
bition of fear-potentiated startle when the reflex was elic-
ited after the offset of S2. Therefore, using this procedure,
second-order fear is not masked by conditioned inhibition,
even after very extensive second-order training. In fact,
rather than an inverted-U function, we found that second-
order conditioning increased monotonically as a function of
the degree of second-order training (Fig. 2).
Higher-Order Conditioning: Window on the
Architecture of Memory
Associative learning involves the development of associa-
tions between neural representations of events (e.g., Ko-
norski 1967). Debate about the nature of these associations
preoccupied traditional learning theorists perhaps more
than any other issue. Much of the argument concerned
whether the organism learned about the relationship be-
tween one stimulus and another (S-S learning), between a
stimulus and a response (S-R learning), or between two
responses (R-R learning). Some theorists attempted to ac-
count for all learning through a single mechanism (e.g., Hull
1943), but it is now more generally appreciated that a con-
ditioned stimulus can enter into an association with a vari-
ety of elements in a given learning situation (Rescorla
1980a; Rashotte 1981). As will be seen below, more than
one form of association can be demonstrated in higher-
order conditioning, depending on the arrangement of S2
and S1. Thus, higher-order conditioning offers a means of
analyzing the structure of associations formed in learning,
thereby providing an important base from which to eluci-
date biological mechanisms of associative learning.
As outlined by Rizley and Rescorla (1972), higher-order
learning might be based primarily on associations between
representations of S2 and S1 (S-S learning), representations
of S2 and the US (S-US learning), or representations of S2
and the conditioned response elicited by the US (S-R learn-
ing; Fig. 3). In sensory preconditioning, the two condi-
tioned stimuli are paired before the US is presented and,
therefore, before the development of a strong conditioned
response to S1. Hence, despite an appeal by some S-R theo-
rists to the existence of covert and unmeasurable behavioral
responses (Miller 1951), sensory preconditioning would
seem to represent as clear a case as one could obtain of S-S
However, it is less easy to predict the nature of the
associations formed in second-order conditioning. These al-
ternatives began to be evaluated in a systematic manner
through a series of elegant experiments (Rizley and Res-
after very extensive training when the fear-potentiated startle para-
digm is used (Gewirtz 1998). There were 16 S2-S1 trials presented
on each day of second-order training. Fear-potentiated startle is
defined as the difference (±SEM) in startle amplitude measured in
the presence versus absence of the CS.
Second-order fear conditioning can still be observed
Higher-Order Conditioning and Substrates of Memory
corla 1972; see also Konorski 1948; Rozeboom 1958). The
general approach was to assess the effects of various post-
training manipulations on the strength of second-order con-
ditioning. For example, if the conditioned response to S1
were extinguished, what would be the effect on the re-
sponse to S2? If S2 is associated with S1, then extinction of
responding to S1 should also cause extinction of responding
to S2. In fact, in sensory preconditioning, the prototypic
measure of S-S learning, repeated nonreinforcement of S1
leads to extinction of responding to S1 as well as extinction
of responding to S2 (e.g., Archer and Sjoden 1982; Rizley
and Rescorla 1972). In contrast, most evidence suggests
that when S2 and S1 are paired in a second-order condition-
ing procedure, nonreinforcement of S1 does not lead to
extinction of responding to S2 (e.g., Rizley and Rescorla
1972). This suggests that S-S associations are not, in general,
the basis of second-order conditioning.
How can one assess the possibility that second-order
retrieval involves direct activation of a representation of the
US (S2-US learning)? Interestingly, this appears to be the
basis of retrieval of first-order associations, as the first-order
conditioned response is sensitive to posttraining alterations
in the value of the US. For example, a CS is paired with food
so that a conditioned response can be measured. The US
can then be devalued by pairing it with sickness. Following
such treatment, animals show a reduction in the condi-
tioned response to the CS compared to animals that did not
experience US devaluation. This suggests that during con-
ditioning, animals form a representation of the value of the
US with which the CS has been paired. When that repre-
sentation is changed (either devalued or inflated), then the
behavior elicited by the cue also is changed in the same
direction. However, second-order conditioning is generally
not affected by these treatments, suggesting that an asso-
ciation between S2 and the US does not constitute the pri-
mary form of learning in second-order conditioning (e.g.,
Holland and Rescorla 1975a; Helmstetter and Fanselow
Further evidence that first- but not second-order con-
ditioning involves the encoding of a detailed representation
of the US was obtained in an elegant study that measured
the topography of pigeons’ keypecks in response to CSs
that had been paired with food or water (Stanhope 1992).
While pigeons pecked more forcefully for an S1 that had
been paired with food than for an S1 that had been paired
with water, the force of keypecks to S2 was not similarly
indicate probable representations activated during retrieval of first-order conditioning (S1-US), second-order conditioning (S2-Fear), and
sensory preconditioning (S2-S1), based on the existing literature. Dotted arrows indicate other possible representations that may be activated
during retrieval of second-order conditioning (see text for explanation).
A schematic representation of hypothetical associations that could be formed in higher-order fear conditioning. Dashed arrows
Gewirtz and Davis
dependent on the nature of the reinforcer with which S1
previously had been paired.
On the basis of the devaluation and inflation data, Res-
corla (1973; 1980a) reasoned that only an association be-
tween S2 and the conditioned response to S1 (S-R) learning
would be immune to changes in the values of S1 and the US.
However, this conclusion was arrived at by default and not
on the basis of any direct evidence. In fact, other evidence
suggests that S2 is not associated with an overt behavioral
response. This is indicated by the finding that the specific
form of the conditioned response produced by S2 may be
different from the conditioned response produced by S1,
when the two stimuli are in different sensory modalities
(Holland 1977; Kim et al. 1996). For example, whereas a
light paired with food produces a rearing response, a tone
paired with the light produces a quick, startle-like response
but little evidence of rearing (Holland 1977).
Evidence of second-order fear-potentiated startle (see
above) provides further demonstration of this point. The
behavior used to measure fear (the startle reflex) is not a CR
evoked by S1 (S1 does not elicit startle but instead increases
the amplitude of the startle reflex elicited by some other
stimulus), and the startle reflex is not elicited at all during
second-order training. Thus, the very existence of second-
order fear-potentiated startle demonstrates that second-or-
der conditioning must involve more than simply an associa-
tion between S2 and one or more overt behavioral re-
sponses produced by S1.
Therefore, although still often referred to in this way,
S-R learning does not really provide an adequate description
of second-order conditioning because S2 is not associated
with a specific overt behavior. The interpretation most con-
sistent with the data set as a whole is that S2 is associated
with a central motivational state (Holland 1977). Thus, in
the case of aversive learning, S2 anticipates a state of fear,
rather than either a representation of the specific US with
which S1 was paired or a specific CR elicited by S1. In the
case of appetitive conditioning, S2 anticipates the delivery
of reward, activating the hedonic system (Robinson and
Berridge 1993) and producing, for wont of a better term, a
state of hope (Mowrer 1960).
It should be noted that the general expectation of re-
ward or punishment does not underlie second-order condi-
tioning under all circumstances. For example, there have
been isolated reports suggesting the presence of S2-US as-
sociations in second-order conditioning (Ross 1986; Barnet
and Miller 1996). In addition, in one conditioning situa-
tion—autoshaping in pigeons—extinction of S1 substan-
tially reduces the conditioned response to S2, suggesting
that S2-S1 associations are the principle basis of second-
order conditioning in this paradigm (Leyland 1977; Rashotte
et al. 1977). This does not reflect simply an idiosyncratic
difference between pigeons and other species because ro-
dents may also acquire S2-S1 second-order conditioning
when two conditions are applied that normally promote
strong sensory preconditioning: when S2 and S1 are pre-
sented simultaneously (as opposed to serially) and when
very few first-order (S1-US) training trials are included (Res-
corla 1982). These conditions minimize the incidence of S2
in the absence of S1 and the incidence of S1 in the absence
of S2. Because the stimuli are not experienced independent
of one another, these arrangements may thus encourage
S2-S1 learning through the formation of a unitary represen-
tation of the two stimuli (Rescorla and Durlach 1981).
In sum, sensory preconditioning involves the associa-
tion between representations of S2 and S1. In contrast, in
second-order conditioning, S2 comes to evoke a general
expectation of reward or punishment, although S2-S1 learn-
ing can be encouraged under specific conditions. In this
respect, it is interesting to note the asymmetry between
second-order conditioning and first-order conditioning, in
which S1 appears to evoke a memory of specific character-
istics of the US. This means that one can use sensory pre-
conditioning, second-order conditioning, and first-order
conditioning to compare the neural substrates of S2-US, S2-
S1, and S2-Fear or S2-Hope learning.
The form of association that predominates in second-
order conditioning may also help in understanding the eti-
ology of certain psychiatric disorders. For example, second-
order conditioning may provide a suitable nonhuman model
for panic disorder, in which symptoms of agoraphobia de-
velop as aversions to places in which panic attacks have
occurred. That is, specific contextual cues may become
associated with the fear of having a panic attack rather than
with whatever triggered the attack in the first place.
Higher-Order Conditioning and the Biology
of Associative Learning and Memory
Mechanisms of Reinforcement I: Fear Conditioning
The foregoing discussion suggests that comparisons of first-
order conditioning, sensory preconditioning, and second-
order conditioning might help us elucidate the neural sub-
strates of three distinct forms of associative learning. The
problem, however, in translating a model of associations
based on behavioral data into a biological model is in refor-
mulating the abstract concepts of S-S, S-US, and S-Fear or
S-Hope learning into neural terms. We have suggested that
these terms be defined with reference to the point or points
of convergence in the brain of pathways conveying CS and
US information (Gewirtz and Davis 1997b). Based on this
idea, we have developed a schematic model of the path-
ways that may be involved in first- and higher-order fear
conditioning (Fig. 4). This model assumes that pathways
conveying S2, S1, and US information converge on one
structure, the amygdala, that constitutes a site of plasticity
of first-order conditioning. The second-order association is
defined as S2-S1 if the S2 pathway inserts into the pathway
Higher-Order Conditioning and Substrates of Memory
mediating first-order conditioning before the point of con-
vergence between S1 and US information. It is defined as
S2-US if it inserts into the first-order pathway at the point of
S1-US convergence; as S2-Fear if it inserts after the point of
S1-US convergence but before the point of divergence to
different response systems; and as S2-R if it inserts after the
point of response system divergence.
We assumed that S1-US convergence required for the
acquisition of first-order fear conditioning occurs in the ba-
solateral complex of the amygdala (BLA), probably in the
lateral nucleus of the amygdala. This assumption is based on
a variety of findings such as that CS and US sensory inputs
converge onto the same cells of the lateral nucleus (Roman-
ski et al. 1993) and that lesions (Campeau and Davis 1995b)
and local infusion of N-methyl-D-aspartate (NMDA) and non-
NMDA-type glutamate receptor antagonists (Miserendino et
al. 1990; Campeau et al. 1992; Kim et al. 1993; Lee and Kim
1998) into the BLA block first-order fear conditioning.
On the basis of the devaluation and inflation data, the
model assumes that expression of first-order conditioning
involves retrieval of a specific memory of the US (i.e., acti-
vation of a representation of the US). We postulated that
this representation is stored outside the amygdala, in the form
of a particular pattern of activity in regions of neocortex acti-
vated by the US itself. Retrieval of the memory for the US
involves the production of a similar pattern of activation in
these structures, initiated by the CS via the amyg-
dala. Although lacking direct support, there are
some data consistent with this view of retrieval
of first-order fearful memories. Lesions made af-
ter training of posterior portions of insular cor-
tex, particularly below the rhinal sulcus (for-
merly defined as part of perirhinal cortex; Rosen
et al. 1992), block expression of first-order con-
ditioning (Rosen et al. 1992; Campeau and Davis
1995a; Corodimas and LeDoux 1995; Falls et al.
1997). In contrast, acquisition is not blocked
when lesions are made before training (Roman-
ski and LeDoux 1992). Interestingly, the critical
locus affected by these lesions contains sparse
projections from visual and auditory areas but
substantial projections from somatosensory cor-
tex. Hence, it is unlikely that this region is re-
quired for transmission of CS information (audi-
tory or visual) to the amygdala (Rosen et al.
1992; Campeau and Davis 1995a) or for retrieval
of contextual information
LeDoux 1995) as has been suggested. Rather,
this region may be part of a network of cortical
structures that, orchestrated by the BLA, stores a
representation of a somatosensory US. Accord-
ing to this view, when lesions are made after
training, the US representation is degraded to the
point where expression of conditioned fear is
blocked; whereas, when lesions are made before training, a
US representation can be acquired adequately using other
cortical areas. Thus, the anatomical evidence and behavioral
data are consistent with the view that lesions disrupt stor-
age and retrieval of the US representation (Gewirtz and
Davis 1997b; Shi 1995). Like first-order conditioning, sec-
ond-order fear conditioning is dependent on NMDA-type
glutamate receptor activation in the BLA (Gewirtz and Davis
1997a). Assuming, then, that second-order conditioning is
similar to NMDA-dependent associative LTP, S2 would rep-
resent the weak stimulus and S1 the strong, depolarizing
stimulus. If so, then one can ask where in this circuitry
would S1 produce strong depolarization that would lead to
S-Fear learning? Presumably this would be somewhere
downstream of the site of first-order (i.e., S-US) plasticity.
For example, within the BLA the lateral nucleus may be the
site of first-order plasticity and the basal nucleus the site of
second-order plasticity. The latter structure receives heavy
projections from both the lateral nucleus (Pitka ¨nen et al.
1995) and from perirhinal cortex (Shi 1995), both of which
would be activated by S1 during second-order training. Re-
cent evidence that might support this view comes from a
study of secondary reinforcement (Amorapanth et al. 2000).
Lesions of the basal nucleus of the amygdala prevented a
fearful CS (i.e., S1) from supporting acquisition of an instru-
mental escape response but did not disrupt freezing to the
higher-order fear conditioning (see text for explanation).
A highly simplified, schematic representation of a circuit for first- and
Gewirtz and Davis
CS itself. In contrast, lesions of the lateral nucleus blocked
both conditioned freezing and the ability of the CS to sup-
port instrumental learning. This might suggest that the plas-
ticity underlying higher-order learning (in this case, CS-sig-
naled escape learning) occurs in the amygdala downstream
of the site of plasticity of first-order conditioning. Alterna-
tively, plasticity underlying higher-order learning may occur
in the lateral nucleus, with the basal nucleus forming a
conduit for expression of escape behavior.
Mechanisms of Reinforcement II: Other Higher-Order
Although the neural substrates of higher-order conditioning
are still relatively uncharted, involvement of the BLA is
emerging as critical in both Pavlovian second-order condi-
tioning and related learning phenomena. First, the BLA is
critically involved in second-order appetitive learning. Ex-
citotoxic lesions of the BLA block acquisition of a second-
order appetitive conditioned response (approach toward a
food cup) but not of the same response conditioned to a
first-order CS (Hatfield et al. 1996). The sparing of the first-
order response rules out an account in terms of perfor-
mance deficits and suggests instead that the BLA is involved
in the acquisition of second-order appetitive conditioning.
In contrast to lesions of the BLA, lesions of the central
nucleus of the amygdala (CeA) block both first- and second-
order conditioned orienting responses but not conditioned
approach to the food (Hatfield et al. 1996). Based on these
and other data, Hatfield et. al. (1996) suggest the CeA “regu-
lates attentional processing of cues during associative con-
ditioning” (Hatfield et al. 1996, p. 5265), whereas the ba-
solateral nucleus of the amygdala is critically involved in
“associative learning processes that give CSs access to the
motivational value of their associated USs” (Hatfield et al.
1996, p. 5264).
In addition, BLA lesions block the effect of US devalu-
ation (Hatfield et al. 1996). Similarly, in instrumental condi-
tioning, lesions of the BLA impair the ability of rats to detect
a decrease in reward magnitude (the “Crespi” or “negative
contrast” effect; Becker et al. 1984; Salinas et al. 1996).
Thus, BLA lesions may block acquisition of second-order
conditioning by blocking access to a memory of the current
motivational value of the reinforcer, stored in the posterior
insular cortex. Importantly, because acquisition and expres-
sion of first-order appetitive conditioning (in contrast to
fear conditioning, see above) are not disrupted by the same
lesions, a memory of the original value of the appetitive
reinforcer must be stored independent of this amygdalo-
As described earlier, lesions of the basal nucleus of the
BLA block the secondary reinforcing effects of a fearful CS
on acquisition of an avoidance response but do not block
freezing (Amorapanth et al. 2000). A similar involvement of
the BLA is indicated in appetitive secondary reinforcement
as well. Lesions of the BLA markedly reduced the ability of
a CS that had been paired with an estrous female to support
instrumental acquisition of a lever response. The animal’s
tendency to respond to the primary reinforcer (the estrous
female) was not impaired (Everitt et al. 1989). Finally, in
another paradigm related to higher-order conditioning, lo-
cal infusion of the NMDA antagonist AP5 into the BLA dis-
rupted acquisition of a taste-potentiated odor aversion but
not acquisition of a first-order taste aversion (Hatfield and
In summary, several findings now suggest a critical role
for the BLA in a variety of higher-order conditioning situa-
tions, both aversive and appetitive.
Combined Analysis of First- and Second-Order
Conditioning to More Precisely Measure
Learning and Memory
Although applying a treatment (e.g., drug, lesion) before
training or testing is typically used to discriminate effects on
learning versus performance, this approach is often still
problematic. Some of the problems can be overcome by
including tests of higher-order conditioning. For example,
consider a treatment, such as a drug infused locally into the
amygdala, that blocks expression of first-order conditioning
when applied before testing, but not acquisition when ap-
plied before training, tested at some later time when the
treatment no longer is in effect (Table 2). Does the treat-
How Second-Order Conditioning Can Help in More Precisely Determining the Effects of a Treatment on Learning
Scenario 1No effectEffectNo effect
Treatment has effect on response expression
Treatment has effect on memory retrieval
Treatment has effect on transmission of US signal
Treatment has effect on acquisition of new memory
Scenario 2 Effect No effect
The treatment is given either before first-order training (to test for an effect on acquisition of first-order conditioning), before first-order
testing (to test for an effect on expression of first-order conditioning), or before second-order training (to test for an effect on acquisition
of second-order conditioning).
Higher-Order Conditioning and Substrates of Memory
ment act by blocking memory activation or does it act fur-
ther downstream by blocking elicitation of a behavioral re-
sponse? To distinguish between these possibilities one can
test the ability of the drug to block acquisition of second-
order conditioning. Recall that S2 becomes associated with
a central state of anticipation of reinforcement rather than
with a specific peripheral response. Therefore, if acquisi-
tion of second-order conditioning also were blocked, it
would strongly suggest that the treatment interfered with
retrieval of the first-order memory and not with perfor-
mance of an overt behavioral response.
This rationale has recently been applied to exploring
effects of dopamine on fear conditioning. Systemic applica-
tion of the D2 agonist quinpirol blocks acquisition of sec-
ond-order conditioning (Nader and LeDoux 1999a). Quin-
pirol does not, however, disrupt acquisition of sensory pre-
conditioning. (The study did not measure the effects of
quinpirol on first-order conditioning, but a sparing of sen-
sory preconditioning effectively provides for the same sorts
of controls.) This suggests that D2 receptor stimulation
does not interfere in a general way with processes involved
in synaptic plasticity or transmission of CS information. The
inability to acquire second-order conditioning suggests that
D2 receptor stimulation (probably in the ventral tegmental
area; Nader and LeDoux 1999b) inhibits the retrieval of a
central fear state, rather than the performance of a fear-
related behavior (freezing).
Second, consider a situation where a drug enhances
acquisition of a first-order association but not its expression
(see Table 2). The most intriguing conclusion would be that
the drug facilitated processes involved in the cellular plas-
ticity that underlies associative learning. Alternatively, the
drug might have acted by enhancing transmission along
central US pathways, perhaps not reflected by lower brain
stem or spinal cord measures of US activation. However, if
the drug also increased the rate of acquisition of second-
order conditioning this could not be interpreted as reflect-
ing alterations in US processing. This is because the drug
would only be given before second-order training sessions,
in which the US was not presented (i.e., no refresher trials
would be included during second-order training). Using this
approach, Cicala et al. (1990) found that the opiate antago-
nist naloxone enhanced second-order conditioning when
given before second-order training. This effect could not be
attributed to a change in shock sensitivity because no
shocks were given after the drug was administered.
We have also used this rationale to further investigate
the role of NMDA-type glutamate receptors in the amygdala
in acquisition of conditioned fear. Local infusion of an
NMDA receptor antagonist into the amygdala at the time of
training disrupts first-order fear conditioning (Miserendino
et al. 1990; Campeau et al. 1992; Fanselow and Kim 1994),
odor-aversion learning (Staubli et al. 1989), appetitive con-
ditioning (Burns et al. 1994; Baldwin et al. 2000), and in-
hibitory avoidance learning (Izquierdo et al. 1992; Kim and
McGaugh 1992; Liang et al. 1994). Expression of first-order
fear conditioning is not disrupted (Miserendino et al. 1990;
Campeau et al. 1992; Liang et al. 1994), indicating that trans-
mission of CS information is intact (but see also Maren et al.
1996; Lee and Kim 1998). Furthermore, intra-amygdala AP5
does not affect reactivity to footshock during fear condi-
tioning (Miserendino et al. 1990; Kim and McGaugh 1992;
Liang et al. 1994). However, this measure may reflect only
a spinally mediated withdrawal reflex. Thus, one cannot
rule out with any confidence the possibility that NMDA
antagonists interfered with local transmission of US infor-
mation in the amygdala during first-order conditioning. To
test whether blockade of NMDA receptors in the amygdala
interferes with associative learning independent of any ef-
fect it might have on US processing, we infused AP5 into
the amygdala immediately before second-order condition-
ing (Gewirtz and Davis 1997a). Importantly, not only did
AP5 completely block second-order conditioning but the
same dose actually enhanced expression of first-order con-
ditioning (Fig. 5). Because first-order fear provides the rein-
blocked by microinfusion of the NMDA antagonist AP5 into the
BLA. The drug was injected immediately before each session of
second-order conditioning. The control group was injected with
artificial cerebrospinal fluid (ACSF) vehicle. The three tests were
conducted before second-order training (pretraining) and after two
(posttraining 1) and three (posttraining 2) sessions of second-order
training. No first-order pairings were given during second-order
training. Fear-potentiated startle is defined as the difference (±SEM)
in startle amplitude measured in the presence versus absence of the
CS. Figure first published in Gewirtz and Davis (1997a) and repro-
duced with permission of the publisher.
Acquisition of second-order fear-potentiated startle is
Gewirtz and Davis
forcement signal for second-order conditioning, this
strongly suggests that AP5 did not block second-order con-
ditioning by disrupting transmission of the reinforcement
signal to the amygdala.
The foregoing discussion has illustrated several ways in
which higher-order conditioning can be used to analyze
neural substrates of learning and memory. The value of
higher-order conditioning stems from the fact that the rein-
forcer has motivational value that is acquired, as opposed to
intrinsic motivational value. As a result, higher-order condi-
tioning can be used to investigate the neural substrates of
different forms of associative memory and to identify pro-
cesses involved in learning and memory versus processes
involved in transmission of sensory or motor information.
Beyond this, however, the value of higher-order condi-
tioning to the neurobiologist or learning theorist may be
greater than its usefulness as a laboratory instrument. It is
likely that most sources of reinforcement in ethological set-
tings have acquired motivational significance rather than
any particular intrinsic value. This point is easy to overlook.
For example, the sight of a banana can clearly serve as a
powerful reinforcing stimulus for a monkey, but the mon-
key first has to learn to associate its appearance with its
odor and taste (Gaffan 1992). Similarly, little human learn-
ing presumably involves the direct pairing of objects or
events with powerful, unconditioned reinforcers. Hence,
higher-order conditioning offers a bridge between Pavlov-
ian conditioning experiments conducted inside the labora-
tory and the types of learning that perhaps predominate in
the outside world.
This work was supported by NIH Grants MH47840, MH57250,
MH58922, MH52384, MH59906, MH11370, and the Woodruff
Amorapanth, P., LeDoux, J.E., and Nader, K. 2000. Different lateral
amygdala outputs mediate reactions and actions elicited by a
fear-arousing stimulus. Nature Neurosci. 3: 74–79.
Archer, T. and Sjoden, P.-O. 1982. Higher-order conditioning and sensory
preconditioning of a taste aversion with an exteroreceptive CS1.
Quart. J. Exp. Psychol. 34B: 1–17.
Baldwin, A.E., Holahan, M.R., Sadeghian, K., and Kelley, A.E. 2000.
N-methyl-D-aspartate receptor-dependent plasticity within a distributed
corticostriatal network mediates appetitive instrumental learning.
Behav. Neurosci. 114: 84–98.
Barnet, R.C. and Miller, R.R. 1996. Second-order excitation mediated by a
backward conditioned inhibitor. J. Exp. Psychol. Anim. Behav.
Process. 22: 279–296.
Becker, H.C., Jarvis, M.F., Wagner, G.C., and Flaherty, C.F. 1984. Medial
and lateral amygdalectomy differentially influences consummatory
negative contrast. Physiol. Behav. 33: 707–712.
Burns, L.H., Everitt, B.J., and Robbins, T.W. (1994). Intra-amygdala infusion
of the N-methyl-D-aspartate receptor antagonist AP5 impairs
acquisition but not performance of discriminated approach to an
appetitive CS. Behav. Neural Biol. 61: 242–250.
Campeau, S. and Davis, M. 1995a. Involvement of subcortical and cortical
afferents to the lateral nucleus of the amygdala in fear conditioning
measured with fear-potentiated startle in rats trained concurrently with
auditory and visual conditioned stimuli. J. Neurosci. 15: 2312–2327.
. 1995b. Involvement of the central nucleus and basolateral
complex of the amygdala in fear conditioning measured with
fear-potentiated startle in rats trained concurrently with auditory and
visual conditioned stimuli. J. Neurosci. 15: 2301–2311.
Campeau, S., Miserendino, M.J.D., and Davis, M. 1992. Intra-amygdala
infusion of N-methyl-D-aspartate receptor antagonist AP5 blocks
acquisition but not expression of fear-potentiated startle to an auditory
conditioned stimulus. Behav. Neurosci. 106: 569–574.
Cicala, G.A., Azorlosa, J.L., Estall, L.B., and Grant, S.J. 1990. Endogenous
opioids interfere with Pavlovian second-order fear conditioning.
Psychol. Sci. 1: 312–314.
Corodimas, K.P. and LeDoux, J.E. 1995. Disruptive effects of posttraining
perirhinal cortex lesions on conditioned fear: Contributions of
contextual cues. Behav. Neurosci. 109: 613–619.
Davis, M., Schlesinger, L.S., and Sorenson, C.A. 1989. Temporal specificity
of fear conditioning: Effects of different conditioned stimulus–
unconditioned stimulus intervals on the fear-potentiated startle effect.
J. Exp. Psychol. Anim. Behav. Process. 15: 295–310.
Everitt, B.J., Cador, M., and Robbins, T.W. 1989. Interactions between the
amygdala and ventral striatum in stimulus-reward associations: Studies
using a second-order schedule of reinforcement. Neuroscience
Falls, W.A., Bakker, K.T., and Heldt, S.A. 1997. Lesions of perirhinal cortex
interfere with conditioned excitation but not with conditioned
inhibition of fear. Behav. Neurosci. 111: 476–486.
Falls, W.A. and Davis, M. 1997. Inhibition of fear-potentiated startle can be
detected after the offset of a feature trained in a serial feature-negative
discrimination. J. Exp. Psychol. Anim. Behav. Process. 23: 3–14.
Fanselow, M.S. and Kim, J.J. 1994. Acquisition of contextual Pavlovian fear
conditioning is blocked by application of an NMDA receptor
antagonist D,L-2-Amino-5-Phosphonovaleric acid to the basolateral
amygdala. Behav. Neurosci. 108: 210–212.
Gaffan, D. 1992. Amygdala and the memory of reward. In The amygdala:
Neurobiological aspects of emotion, memory, and mental
dysfunction (ed. J.P. Aggleton). Wiley-Liss, New York.
Gewirtz, J.C. 1998. Second-order conditioning of fear-potentiated startle.
Unpublished Ph.D. thesis, Yale University.
Gewirtz, J.C. and Davis, M. 1997a. Second-order fear conditioning
prevented by blocking NMDA receptors in amygdala. Nature
. 1997b. Application of Pavlovian higher-order conditioning to the
analysis of the neural substrates of fear conditioning.
Neuropharmacology 37: 453–459.
Hatfield, T. and Gallagher, M. 1995. Taste-potentiated odor conditioning:
Impairment produced by infusion of an N-Methyl-D-Aspartate
antagonist into basolateral amygdala. Behav. Neurosci. 109: 663–668.
Hatfield, T., Han, J.-S., Conley, M., Gallagher, M., and Holland, P. 1996.
Neurotoxic lesions of basolateral, but not central, amygdala interfere
with Pavlovian second-order conditioning and reinforcer devaluation
effects. J. Neurosci. 16: 5256–5265.
Helmstetter, F.J. and Fanselow, M.S. 1989. Differential second-order
aversive conditioning using contextual stimuli. Anim. Learn. Behav.
Herendeen, D. and Anderson, D.C. 1968. Dual effects of a second-order
conditioned stimulus: Excitation and inhibition. Psychon. Sci.
Holland, P.C. 1977. Conditioned stimulus as a determinant of the form of
the Pavlovian conditioned response. J. Exp. Psychol. Anim. Behav.
Process. 3: 77–104.
Holland, P.C. and Rescorla, R.A. 1975a. The effect of two ways of
devaluing the unconditioned stimulus after first- and second-order
appetitive conditioning. J. Exp. Psychol. Anim. Behav. Process.
Higher-Order Conditioning and Substrates of Memory
. 1975b. Second-order conditioning with food unconditioned
stimulus. J. Comp. Physiol. Psychol. 88: 459–467.
Hull, C.L. 1943. Principles of behavior. Appleton-Century-Crofts, New
Izquierdo, I., Da Cunha, C., Rosat, R., Jerusalinsky, D., Ferreira, M.B.C.,
and Medina, J.H. (1992). Neurotransmitter receptors involved in
memory processing by the amygdala, medial septum and hippocampus
of rats. Behav. Neural Biol. 60: 16–26.
Kamil, A.C. 1969. Some parameters of the second-order conditioning of
fear in rats. J. Comp. Physiol. Psychol. 67: 363–369.
Kehoe, E.J., Feyer, A.-M., and Moses, J.L. 1981. Second-order conditioning
of the rabbit’s nictitating membrane response as a function of the
CS2-CS1 and CS1-US intervals. Anim. Learn. Behav. 9: 304–315.
Kim, M. and McGaugh, J.L. 1992. Effects of intra-amygdala injections of
NMDA receptor antagonists on acquisition and retention of inhibitory
avoidance. Brain Res. 585: 35–48.
Kim, M., Campeau, S., Falls, W.A., and Davis, M. 1993. Infusion of the
non-NMDA receptor antagonist CNQX into the amygdala blocks
expression of fear-potentiated startle. Behav. Neural Biol. 59: 5–8.
Kim, S.D., Rivers, S., Bevins, R.A., and Ayres, J.J.B. 1996. Conditioned
stimulus determinants of conditioned response form in Pavlovian fear
conditioning. J. Exp. Psychol. Anim. Behav. Process. 22: 87–104.
Konorski, J. 1948. Conditioned reflexes and neuron organization. Hafner,
Konorski, J. 1967. Integrative activity of the brain. University of Chicago
Lee, H. and Kim, J.J. 1998. Amygdalar NMDA receptors are critical for new
fear learning in previously fear-conditioned rats. J. Neurosci.
Leyland, C.M. 1977. Higher order autoshaping. Quart. J. Exp. Psychol.
Liang, K.C., Hon, W., and Davis, M. 1994. Pre- and posttraining infusion of
N-Methyl-D-Aspartate receptor antagonists into the amygdala impair
memory in an inhibitory avoidance task. Behav. Neurosci.
Maren, S., Aharonov, G., Stote, D.L., and Fanselow, M.S. 1996.
N-Methyl-D-Aspartate receptors in the basolateral amygdala are
required for both acquisition and expression of conditioned fear in
rats. Behav. Neurosci. 110: 1365–1374.
McAllister, D.E. and McAllister, W.R. 1964. Second-order conditioning of
fear. Psychon. Sci. 1: 383–384.
Miller, N.E. 1951. Comments on multiple-process conceptions of learning.
Psychol. Rev. 58: 375–381.
Miserendino, M.J.D., Sananes, C.B., Melia, K.R., and Davis, M. 1990.
Blocking of acquisition but not expression of conditioned
fear-potentiated startle by NMDA antagonists in the amygdala. Nature
Mowrer, O.H. 1960. Learning theory and symbolic processes. Wiley, New
Nader, K. and LeDoux, J.E. 1999a. The dopaminergic modulation of fear:
Quinpirole impairs the recall of emotional memories in rats. Behav.
Neurosci. 113: 152–165.
. 1999b. Inhibition of the mesoamygdala dopaminergic pathway
impairs the retrieval of conditioned fear associations. Behav. Neurosci.
Pavlov, I.P. 1927. Conditioned reflexes: An investigation of the
physiological activity of the cerebral cortex (G.V. Anrep, trans.).
Dover, New York.
Pfautz, P.L., Donegan, N.H., and Wagner, A.R. 1978. Sensory
preconditioning versus protection from habituation. J. Exp. Psychol.
Anim. Behav. Process. 4: 286–295.
Pitka ¨nen, A., Stefanacci, L., Farb, C.R., Go, G.G., LeDoux, J.E., and Amaral,
D.G. 1995. Intrinsic connections of the rat amygdaloid complex:
Projections originating in the lateral nucleus. J. Comp. Neurol.
Prewitt, E.P. 1967. Number of preconditioning trials in sensory
preconditioning using CER training. J. Comp. Physiol. Psychol.
Rashotte, M.E. 1981. Second-order conditioning: Contributions to the
research and theory of Pavlovian reinforcement by conditioned
stimuli. In Autoshaping and conditioning theory (eds. C.M. Locurto,
H.S. Terrace, and J. Gibbon). Academic Press, New York.
Rashotte, M.E., Griffin, R.W., and Sisk, C.L. 1977. Second-order
conditioning of the pigeon’s keypeck. Anim. Learn. Behav. 5: 25–38.
Rescorla, R.A. 1973. Second-order conditioning: Implications for theories
of learning. In Contemporary approaches to conditioning and
learning (eds. F.J. McGuigan and D.B. Lumsden), pp. 127–150. V.H.
Winston, Washington DC.
. 1979. Aspects of the reinforcer learned in second-order Pavlovian
conditioning. J. Exp. Psychol. Anim. Behav. Process. 5: 79–95.
. 1980a. Pavlovian second-order conditioning: Studies in
associative learning. Erlbaum, NJ.
.1980b. Simultaneous and successive associations in sensory
preconditioning. J. Exp. Psychol. Anim. Behav. Process. 6: 207–216.
. 1982. Simultaneous second-order conditioning produces S-S
learning in conditioned suppression. J. Exp. Psychol. Anim. Behav.
Process. 8: 23–32.
. 1999. Partial reinforcement reduces the associative change
produced by nonreinforcement. J. Exp. Psychol. Anim. Behav. Process.
Rescorla, R.A. and Cunningham, C.L. 1979. Spatial contiguity facilitates
Pavlovian second-order conditioning. J. Exp. Psychol. Anim. Behav.
Process. 5: 152–161.
Rescorla, R.A. and Durlach, P.J. 1981. Within-event learning in Pavlovian
conditioning. In Information processing in animals: Memory
mechanisms (eds. N.E. Spear and R.R. Miller). Erlbaum, NJ.
Rescorla, R.A., and Furrow, D.R. 1977. Stimulus similarity as a determinant
of Pavlovian conditioning. J. Exp. Psychol. Anim. Behav. Process.
Rizley, R.C. and Rescorla, R.A. 1972. Associations in second-order
conditioning and sensory preconditioning. J. Comp. Physiol. Psychol.
Robinson, T.E. and Berridge, K.C. 1993. The neural basis of drug craving:
An incentive-sensitization theory of addiction. Brain Res. Rev.
Romanski, L.M. and LeDoux, J.E. 1992. Bilateral destruction of neocortical
and perirhinal projection targets of the acoustic thalamus does not
disrupt auditory fear conditioning. Neurosci. Lett. 142: 228–232.
Romanski, L.M., Clugnet, M.-C., Bordi, F., and LeDoux, J.E. 1993.
Somatosensory and auditory convergence in the lateral nucleus of the
amygdala. Behav. Neurosci. 107: 444–450.
Rosen, J.B., Hitchcock, J.M., Miserendino, M.J.D., Falls, W.A., Campeau, S.,
and Davis, M. 1992. Lesions of the perirhinal cortex but not of the
frontal, medial, prefrontal, visual, or insular cortex block
fear-potentiated startle using a visual conditioned stimulus. J. Neurosci.
Ross, R.T. 1986. Pavlovian second-order conditioned analgesia. J. Exp.
Psychol. Anim. Behav. Process. 12: 32–39.
Rozeboom, W.W. 1958. “What is learned?”—An empirical enigma. Psychol.
Rev. 65: 22–33.
Salinas, J.A., Parent, M.B., and McGaugh, J.L. 1996. Ibotenic acid lesions of
the amygdala basolateral complex or central nucleus differentially
effect the response to reductions in reward. Brain Res. 742: 283–293.
Shi, C. 1995. The anatomical substrates underlying the role of the
amygdala in associative learning. Unpublished Ph.D. thesis, University
Stanhope, K.J. 1992. The representation of the reinforcer and the force of
the pigeon’s keypeck in first- and second-order conditioning. Quart. J.
Exp. Psychol. 44: 137–158.
Staubli, U., Thibault, O., DiLorenzo, M., and Lynch, G. 1989. Antagonism
of NMDA receptors impairs acquisition but not retention of olfactory
memory. Behav. Neurosci. 103: 54–60.
Tait, R.W., Marquis, H.A., Williams, R., Weinstein, L., and Suboski, M.D.
1969. Extinction of sensory preconditioning using CER training. J.
Comp. Physiol. Psychol. 69: 170–172.
Yin, H., Barnet, R.C., and Miller, R.R. 1994. Second-order conditioning and
Pavlovian conditioned inhibition: Operational similarities and
differences. J. Exp. Psychol. Anim. Behav. Process. 20: 419–428.
Gewirtz and Davis