Performance of Betta
in a radial
R. L. ROITBLAT, WILLIAM TRAM, and LEONARD GOLUB
New York, New York 10027
were tested in aquatic versions of radial arm mazes. In
the first experiment, the fish were trained to find tubifex worms in an eight-arm maze in which
the optimal strategy was to chooseeach arm once without repetition. After initial training, the
fish entered approximately 6.63 different arms in eight choices, showing a strong tendency to
choose sequences of adjacent arms, moving about the maze in a stereotypic direction. This al-
gorithmic response pattern was not, however, sufficient to predict the high performance level
of the fish. In the secondexperiment, a delay of .5 or 5 min was interposed between the fourth
and fifth choices. Similar stereotypic patterns continued in Experiment 2,
following the longer delay declinedto a levelnot significantly above chance. In the third experi-
ment, different fish weretested in a three-arm maze,reinforcedeither for returning from the sec-
ond arm to the arm in which they had most recently been fed (win-stay) or for visiting a third
arm (win-shift). The fish were significantly faster at acquiring the win-shift contingency than
the win-stay contingency. These results demonstrate
solution of spatial tasks depends on
the interaction ofappropriate behavioral strategies and cognitive capacities
may have little
generality across species.
The importance of a memory for places (spatial
memory) to an animal's survival has often been noted.
For example, nectar feeding birds such as the amakihi
(Loxops virensi studied by Kamil (1978) appear to
use spatial memory to improve their foraging effi-
ciency. Within their own territory, they were less
likely than would be expected by chance to return to
flower clusters on which they had previously fed.
Amakihi intruding on the territory under observa-
tion, however, neither affected nor were affected by
the distribution of flowers on which the resident
birds fed, thereby ruling out any role for cues pro-
duced by the presence of nectar or by the visitation
of an amakihi. Other means for avoiding repeat vis-
its, such as stereotypical search patterns, were also
Similar memory abilities have also been found in
rats performing in a radial arm maze in which each
of the arms extends from a common central platform
like spokes from a wheel. Typically, a single piece of
food is placed at the end of each arm at the start of a
trial and the animal is allowed to enter a number of
arms. Once an arm is visited, no more food is avail-
able there, so the optimal strategy is to enter each
arm exactly once before repeating any choices.
When trained on a four-arm maze, rats perform
This research was supported in part by Grant BNS
from the National ScienceFoundation. Reprint request should be
directed to H.
L. Roitblat, Department of Psychology, 406
Schermerhorn Hall, Columbia University, New York, New York
10027. We thank Robert Scopatz for his careful reading of the
manuscript and his helpful comments.
perfectly after about 10 training trials (Olton, 1978).
In a maze with 17 arms, rats reach an asymptotic
level of choosing about
15 arms in the first 17 choices
50 tests (Olton, Collison, & Werz, 1977).
Like the amakihi, rats' performance on radial arm
mazes appears to be memory based. In one experi-
ment, for example, the arms
the maze were rotated
to new spatial locations and rebaited after the third
choice. The rats continued to enter arms in previ-
spatial locations even if those arms
(in their former locations) had been previously de-
& Samuelson, 1976), thus indicating
that external cues emanating from the food itself
(e.g., odor) or from the passageof a rat (scentmarkers
left on the maze) played a negligible role in control-
ling performance (see also Olton
& Collison, 1979).
Also similar to the amakihi, rats' performance is
not usually dependent on consistent response pat-
terns. Although there was enough regularity in re-
sponse patterns to reject the hypothesis that the rats
were choosing randomly among the arms without re-
1978), the regularities were not
sufficient to account for the high accuracy obtained.
Furthermore, when rats were confined to the central
platform of a
17-arm maze for 20 sec following each
choice, performance remained high while the regu-
larity virtually disappeared (Olton, Collison,
1977). Finally, performance remained high when the
first four choices were forced to arms chosen by the
experimenter, leaving the last four as free choices
& Dale, 1981). These
rat radial-arm-maze performance is
1982 Psychonomic Society,Inc.
20 25 30 35 40 45 50 55
Figure 1. Performance of Betta
in an eigbt-arm radial
maze during acquisition, Experiment 1. Cboice accuracy refers to
tbe number of different arms entered in tbe first cboices. Chance
is the value that would
be expected if the fIsh were sampling ran-
domly witb replacement from among the arms (5.23).
based on internal representations
tion (Roitblat, in press a).
Similar memory capacities have also been reported
jumping goby (Bathygobius soporator)
by Aaronson (1951). These fish swim over a tidal
area during high tide and apparently learn the topog-
raphy of the substrate. At low tide levels, they jump
from one tide pool to another even though unable to
see neighboring pools. They rarely, if ever, jump
onto dry land, passing through as many as six or
seven pools while jumping over ledges and rocks on
their way to the open sea.
In the present study, Siamese fighting fish
were tested for their ability to perform a
spatial task based on the radial arm maze. The in-
dividuals typically maintained by aquarists were orig-
inally derived from wild stock living in Southeast
Asia in various habitats, including slow-moving
streams and other aquatic environments in which
their favorite food, mosquito larvae, can be found
is displayed in Figure 1. Chance is the level predicted
on the basis
random sampling with replacement
The fish showed a strong tendency to choose se-
adjacent arms, moving about the maze in
a singledirection, clockwise for some fish and counter-
clockwise for the others. This tendency is illustrated
in Figure 2 in the form of the percentage of choices
at each displacement. The arm entered on any partic-
ular choice was designated arm
The adjacent arm
in the clockwise direction was called displacement
+I, the second arm
etc. Similarly, the adjacent
arm in the counterclockwise direction was designated
etc. Displacement 4 in either direc-
tion was the arm directly opposite to the chosen arm.
Subjects. Six experimentally naive Betta splendens, three males
and three females, approximately 6 months of age at the start of
the experiment, served as subjects. They were maintained in male-
female pairs in separate compartments of a standard 30-gal tank.
The floor of the tank was covered with 5-8 em of aquarium gravel.
Water was constantly filtered and maintained between 21° and
28°C at pH 7. Approximately one-third of the water in the tank
was changed every other week. During a pretest, the fish were
given access to a large number of tubifex worms, of which they
consumed 20-25within an approximately
Apparatus. An eight-arm maze was constructed of .3-cm while
Plexiglas. Each arm was 5 em wide, 23 em long, and 15 cm high,
equally spaced around a central platform 14.5 cm in diameter.
The end of each arm opposite to the central platform was closed by
fine nylon mesh. A plastic feeding cup (5.08 cm in diameter and
.64 em high) was attached to the floor of each arm 1.25 ern from
the distal end and equidistant from the sides. The maze was sub-
merged in a large Plexiglas tank filled to a depth of 10 em. A
motorized filter forced water thrugh a diffused opening in the
center of the central platform, creating a flow of water outward
from the center toward the arms. The entire apparatus was il-
luminated by overhead incandescent and fluorescent lamps. Maze
temperature was maintained at about the same level as in the home
tank. The water was changed approximately every other week.
Procedure. The fish were transferred to the maze in a small net
and confined to the central area for approximately 2 min in a
white cylinder, 11 em in diameter and 15 em high, while a single
live tubifex worm was placed in the feeding cup at the end of each
arm. The cylinder was then raised and the fish were allowed to
make eight choices (arm entries). A choice was counted when the
base of the fish's tail passed the entrance to the arm. Each fish was
tested twice daily, with a minimum of
IS min intervening between
trials, for a total of 55 trials.
x----x 5 Min.
-4 -3 -2 -I 0
+2 +3 +4
After the first few trials, all fish were entering the
arms and readily consuming the bait. During the first
five trials, the fish chose an average
5.70 arms in
the first eight choices. The accuracy
Figure 2. Proportion of arms chosen at each distance and direc-
tion from tbe current arm. Data for each curve are averages of all
fisb during the last 20 sessions of Experiment
1 and tbe two phases
of Experiment 2.
in the positive direction were removed
from the current arm in tbe clockwisedirection. Negative displace-
ments are in the counterclockwise direction.
BF2X BS3 BF3 BS4 BF4 BS2
vious choices in making its next choice. Figure 3
shows that the observed performance of the real fish
was always superior to that obtained from the sim-
ulated no-memory fish [paired comparison t(5)
12.14, p < .01]. The average choice accuracy of the
simulated fish was 5.82, not significantly above the
chance accuracy of 5.23 [t(5)
=1.57, P > .05].
The probability of repeating a previously chosen
arm (making an error) as a function if the ordinal
position of that choice was also determined for the
live fish. After considering the differing number of
opportunities available, errors were no more likely to
arms visited early in the trial than to arms visited
later in the trial. SeeTable 1.
Figure 3. Average number of different arms entered by fisb
during tbe first eigbt cboices of tbe last 20 trials of Experiment 1
and tbe number expected from a simulation based on tbe assump-
tion tbat tbe fisb were using only transition probabilities and no
memory to determine tbeir radial maze performance.
A new arm 0 was designated on each choice, so the
figure displays choices relative to the immediately
previous choice rather than in terms of absolute
Clockwise turns accounted for
of all choices
made. Choices of displacements
for 63.6% and 17.1% of the choices, respectively.
To evaluate the effect of turning bias on the obtained
choice accuracy, we performed a Monte Carlo sim-
ulation. Choice distributions analogous to Figure 2
were computed separately for each fish during the
last 20 trials of the experiment. These choice prob-
abilities were then used to determine the next choice
in the simulated sequence without any contribution
from previous choices. That is, we assumed in the
simulation that the fish used no memory of any pre-
Conditional Choice Frequencies in Experiment 1
Frequency of Revisit to Arm Most
Recently Visited on Choice
2 0 0
13 0 13
4 1 11 0 12
13 0 27
3 10 0
25 22 15
5 4 3 2
5 5.5 5
opportunities to revisit that position. The
row corresponds to the choice on which the erroroccurred;the
column corresponds to the choice on which that arm was last
The performance of the fish in this experiment
showed a strong algorithmic component that played
a role in determining the direction and the displace-
ment of the sequential choices. This algorithmic com-
ponent was not sufficient, however, to account for
all of the behavior observed. Simulations based only
on the algorithmic component did not differ from an
expectation based on random choice with replace-
ment. Real fish, however, differed not only from this
expectation, but also from the performance of the
simulated fish. Because the simulation was based on
the performance of individual fish, the difference
between simulated and live fish could not have been
due to averaging several fish together and so distort-
ing the algorithm for
fish. The simulation
did combine data from the same fish obtained on 20
the fish were using one algorithm during
one session and another algorithm on a second ses-
sion, both consistent within a day, but different across
days, then the model would predict lower perfor-
mance than would be obtained. No solution is avail-
able for this difficulty using these techniques, how-
The difference between the simulated and the real
fish suggests that some small amount of memory was
involved in determining the real fish's choice. The
most obvious kind of memory that might be used
would be a list (e.g., a pushdown stack) in which each
arm is placed on the list as it is entered, Partial mem-
ory would correspond to maintenance of only the
most recent part of the list. Arms on the list would be
avoided, and arms not on the list might be entered.
No difference was found in the likelihood to reenter
an arm as a function of serial position, meaning either
that the list was very short or that performance was
controlled not by a memory for the arms already vis-
ited, but by a memory for the arms remaining with
food (see Roitblat, in press a). Alternatively, this pat-
tern of choices may represent strategic choices rather
than memory limitations. That is, even though the
fish could remember more information, their pre-
BETTAS IN A RADIALMAZE 111
.5 5 0 .5 5
movement in the maze
resulted in the pat-
These results are similar to those obtained with
& Lamb, 1981) in a radial
maze. Pigeons entered an average
arms in their first eight choices, a value significantly
expected by chance
by a simulation
similar to ours based on
bias. The best predictor
pigeons' performance was based on the assump-
tion that the birds were remembering a maximum
their most recent choices.
Brief confinement following each choice (Olton,
& Werz, 1977) or a single, longer confine-
ment between choices 4
5 (Maki, Brokofsky, &
disrupt the regularities
fail to disrupt choice accuracy.
In contrast, the fish in Experiment 1 showed stronger
did the rats. Accurate
algorithm-based performance depends on an un-
responses; any interruption
algorithm would reset it.
resume the algorithm from the most
recent choice because it does not
should be great if performance depends only on an
if performance has a memory component. Experi-
ment 2 was conducted to assess the effects
Subjects and Apparatus. The subjects and the apparatus were
the same as in the first experiment. One fish died before com-
pleting Phase 2 of the experiment.
Procedure. Experiment 2 was conducted in two phases, each
involvinga delay interpolated betweenthe fourth and fifth choices.
During the delay, the fish were confined to the central platform
inside the "holding" cylinder. During the 20 trials of Phase I,
the fish were confined for 30 sec following the fourth choice. Dur-
ing the 20 trials of Phase 2, the duration of the confinement was
increased to 5 min. In all other respects, the procedure was the
same as that used in the first experiment. The arms were not re-
baited after the start of a trial.
Confinement clearly interfered with accurate per-
formance. Overall choice accuracy (number
ferent arms entered during the first eight choices)
declined with increasing delay [a repeated-measures
ANOVA with days nested in delay condition yielded
F(2,8) = 4.50, P
< .05, for delay). An average
different arms were entered on each
the last 20
during Phase 1
during Phase 2
Experiment 2 (both pre-
confinement choices are included). Choice accuracy
on trials including the
confinement was sig-
Figure 4. Tbe number of novel arms entered on tbe first (left
panel) and on tbe last (rigbt panel) four cboices of eacb trial in Ex-
periments 1(last 20 trials) and 2. Novelarms refer to tbose not pre-
viously entered in tbat trial (i.e., including botb tbose entered in
tbe first and on tbe last four cboices).
nificantly above chance [t(5)
7.60, P < .(01),
accuracy on trials involving a 5-min delay was
2.00, p > .05).
Performance during the first four choices was next
analyzed separately from the performance for the
last four choices (postconfinement in Experiment 2).
the delay was mainly to interfere with
the algorithm during the delay.
the first four choices, displayed
in the left panel
Figure 4, was not significantly dif-
ferent in either delay condition from that observed in
Experiment 1 (both t values less
1). These re-
sults were not surprising, since the first four choices
were made under similar conditions for the three
phases. Performance on the last four choices, seen
in the right panel
Figure 4, did
relative to that obtained with no delay [Phase 1,
<.05; Phase 2, t(4)= 3.83, P < .05). Re-
reentries to arms visited before the delay are
also counted as errors.
During the confinement period, the fish were ac-
the holding cylinder.
at each displacement, was
entered on the fifth choice is displayed in Figure 5,
relative to the
chosen on the fourth choice. The
obtained distributions differed from a uniform dis-
tribution in which each
is chosen with equal fre-
quency when no delay
(x2= 211.958, p < .(05)
(x2= 22.667, P < .005) occurred,
not when a 5-min delay (x2=7.594, P > .05) inter-
vened between the fourth
fifth choices. These
the fish were maintaining
their fourth choices for some time
during the delay. We do not know, however, how
that information was represented (e.g., memory,
postural attitude, etc.; see Roitblat, in press a).
-I 0 +1 +2 +3 +4
Figure 5. The distribution of
of fiftb choices relative
to the arm chosen on the fourth choice. Displayed are the data
from the last 20 trials of Experiment 1 and from the two phases
of Experiment 1. The format is analogous to that of Figure 1.
Whatever role memory played in controlling per-
formance in the first experiment, it was effectively
removed by the longer confinement condition. As
Figure 2 shows, the fish continued to choose adja-
cent arms during Experiment 2, but during the con-
finement the fish were apparently increasingly likely
to "lose their place,"
so to choose randomly
(compare the distributions
fifth choices as a func-
confinement duration, Figure 5). These re-
sults confirm the important role played by a combi-
nation of memory and algorithmic response strategies
in the control
fish spatial performance. The issue
strategic control was further pursued in Experi-
Rat and fish spatial memory performance is con-
trolled by a combination of memory ability and com-
patible response strategy. In the radial arm maze,
animals are reinforced for adopting a win-shift strat-
egy; once food has been obtained in a particular lo-
cation, no more is available there, so the best strategy
is to search elsewhere. Rats seem to prefer discrim-
ination problems that are compatible with this strat-
egy to situations involving a win-stay contingency.
In win-stay situations, reinforcers are delivered in
consistent locations, so once food is found in a par-
ticular location, it pays to return there.
rats' preference for win-
shift situations are available, including spontaneous
alternation in aT-maze (Dember & Fowler,
avoidance of locations already visited in a complex
maze constructed of three concentric hexagonal al-
leys interconnected by shorter alleys (Battig, Driscoll,
& Ulster, 1976; Uster, Battig, & Nageli,
1976) and in the "towers of Baltimore" problem
(Olton, Walker, Gage,
& Johnson, 1977), in which
rats were tested in a large enclosure containing three
towers. On any given day, one
the towers, chosen
randomly, had a number of food pellets placed on
it; the other towers were empty. The rat could
not tell, without climbing a tower whether that tower
had food on that day. Instead of following the op-
searching the towers until food was
discovered and then removing the food from that
tower, the rats visited each tower each day, even after
they had taken some of the food from the single tower
that had any. In fact, the probability
tower with food remained at chance level until the
other two towers had been visited. Thus, in this situa-
tion, the rats were employing a win-shift as well as a
Rats also learn a discrimination task more easily
when the task requires a win-shift as opposed to a
win-stay strategy. Rats have been tested (Olton, in
press; Olton, Handelmann,
& Walker, 1981) in a
the three-table problem introduced by
Maier (1932). At the beginning
each trial, the rats
were fed a small amount
food on one
terconnected tables. A different table was chosen
each day in random order. After a delay, spent in its
home cage, the animal was placed on one
two tables and allowed to choose between the orig-
inal table on which it had been fed and the third table.
The rats that were reinforced for choosing the third
table (whose location also varied from day to day)
learned to perform this task much more quickly
did the rats reinforced for returning
to the table on which they had originally been fed
(Olton, in press).
Although the fish in Experiments 1 and 2 relied
heavily on algorithmic strategies, and only lightly on
memory, to perform the task, it is possible that if we
had asked the question differently, the fish would
have shown the same magnitude of memory seen in
example, rats required to return to the same
arms of an eight-arm maze from which they had re-
cently obtained food perform very poorly (Olton, in
press), while rats required to visit different arms per-
form very well.
the rats' ability to remember the
arms depleted of food in one context were not known,
it might be possible to mistake the poor performance
in the other for a lack of memory.
we ran another group of
on an aquatic
version of the three-table problem.
preferences present in fish are different from those
in rats, then we would expect to see faster acquisition
of a win-stay
of a win-shift discrimination. On
the other hand, the fish could be indifferent to the
strategic aspects of the task, perhaps because these
strategic aspects evolve only in conjunction with a
well-developed spatial memory capacity.
Subjects. The subjects were 12 experimentally naive female
Siamese fighting fish, otherwise similar to those serving in the first
two experiments, maintained under the same conditions.
Apparatus. An aquatic version of the three-table apparatus was
constructed of white Plexiglas. Three arms were arranged in a Y
shape around a triangular central compartment, 7.6 cm per side.
Each arm was 23 cm long, 15 cm high, and 7.6 em wide. The dis-
tal end of each arm was terminated by nylon mesh. A food cup,
2.54 em in diameter and .6 em high, was fastened 1.25
the rear of each arm. Guillotine doors were placed at the entrance
and 10 ern from the rear of each arm, dividing the arm into a
compartment. In order to ensure that
the fish were able to discriminate the three arms, colored plastic
forms were attached to the floor of each arm. One arm had red
(S cm diam), one arm had black squares (4 x 4 em), and
the third had yellow triangles (3 cm per side). Each arm had five
forms equally spaced. The apparatus was submerged in the same
tank and aerated and illuminated as in the first two experiments.
Procedure. At the start of each trial, each arm was randomly
The fish were con-
fined (by means of the guillotine door at the
the placement arm and allowed to consume a single tubifex worm
with which the arm had been prebaited. The fish were then trans-
ferred by net to the start arm and confined in the holding com-
partment. Confinement in the start arm was sufficient to allow
(re)baiting of one of the other two arms and never longer than
S sec. Following this confinement period, the fish were released
and allowed to choose one of the two available arms: the place-
ment arm or the goal arm.
The fish were divided into two experimental squads of six fish
each. Squad I was exposed to a win-stay contingency in which the
placement arm was rebaited during the confinement in the start
arm and the animals were reinforced for returning to that arm.
Squad 2 was exposed to a win-shift contingency in which the goal
arm was baited during confinement in the start arm and the fish
were reinforced for visiting the novel arm. Animals in each squad
were tested on the designated contingency until reaching a criterion
of 8 choices of the baited arm in 10 trials or for a maximum of 70
trials. On reaching criterion, each fish was transferred to the oppo-
site task (i.e., from win-shift to win-stay, or vice versa).
the fish died during the course
periment, one from Squad A
one from Squad B.
from the remaining fish are displayed in
Table 2. The first column
the table shows the total
Number of Trials to Reach Criterion in the Three-Arm Task
Note-Fish in Group A wcre rested first in the shift condition
in the stay condition. Fishin Group 8 were tested first
in the stay condition and then in the shift condition. A blank
indicates that the fish died before completing that condition.
trials each animal received under the win-
shift contingency; the second column shows the total
trials each animal received under the win-
stay contingency. Fewer trials were necessary to reach
criterion in the win-shift condition than in the win-
stay condition (Mann-Whitney U
p < .01).
Under the shift contingency, 9
reached criterion, while under the stay contingency,
10animals reached criterion.
Similar to rats,
Betta splendens also appear to pre-
fer win-shift over win-stay strategies, learning the
former more readily. Similar win-shift preferences
have also been seen in several other species, including
Hawaiian honeycreepers (Kamil, 1978)
(Bombus sonorus) foraging for nectar from
flowers (Witham, 1977).
On the other hand, at least two species
foraging for earthworms and other small terrestrial
invertebrates have exhibited win-stay strategies in un-
constrained (at least by the experimenter) situations.
Smith (1974a, 1974b) has described the pattern em-
ployed by English thrushes searching in a meadow.
On finding prey, the birds increased their rate
decreased their rate
forward motion, stay-
ing in the vicinity
the discovered prey. Similar per-
formance has been obtained from ovenbirds
when searching for mealworms hidden
in a large enclosure on a forest floor (Zach & Falls,
1976). Although the situations in these studies are not
precisely comparable to the three-table problem, they
suggest that employment
a win-shift strategy is
unlikely to be an inherent property
Numerous approaches are available to explain the
differences in performance
across species. Some
these approaches favor some
phylogenetic scale, perhaps based on neural
complexity or phylogenetic ancestry (e.g., Bitterman,
1965). More recent approaches (e.g., Roitblat, in
press b) emphasize the evolution of cognitive capaci-
ties in the context
the animal's ecology. According
to this latter view, each species' cognitive capacities
evolve as adaptive specializations relative to the spe-
cies' ecological niche. One
the questions inherent
in the adoption
concerns the capacities
strategies that co-occur
either because each is necessary for the other to be
benefit (e.g., because one is a precursor or com-
the other) or because they are both a pro-
the same mechanism.
is clear from the
experiments that a high level
formance in a radial arm maze does not depend on
a high-capacity memory system,
rather that performance depends on the interaction
of strategic preferences and memory. That these fish
demonstrate little use of working memory, but com-
pensate through the use of strong algorithmic pat-
terning suggests that memory and strategic prefer-
ences are complementary and are not necessary con-
comitants of one another.
L. R. Orientation and jumping behavior in the gobiid
fish, Bathygobius soporator. American Museum Novitiates,
Effects of nicotine on the exploratory locomotion patterns of
female Roman high- and low-avoidance rats.
Biochemistry and Behavior,
M. E. Phyletic differences in learning. American
R. G., &
M. R. Spatial memory and
the performance of rats and pigeons in the radial arm maze.
W. N., &
H. Spontaneous alternation be-
Psychological Bulletin, 1958,55,412-428.
A. C. Systematic foraging by a nectar-feeding bird, Loxops
Comparative and Physiological Psychology,
MAIER, N. R. F. The effect
cerebral destruction on reasoning
learning in rats. Journal
Comparative Psychology, 1932,
MAKI, W. S.,
S., & BERG, B. Spatial memory in
rats: Resistance to retroactive interference.
D. S. Characteristics of spatial memory. In S. H. Hulse,
& W. K. Honig (Eds.), Cognitive processes in animal
D. S. Staying and shifting: Their effects on discrimination
learning. In M. I. Commons, R. J. Herrnstein,
& H. Rachlin
behavior. Cambridge, Mass:
Ballinger, in press.
D. S., &
C. Intramaze cues and
fail to direct choice behavior on an elevated maze.
Learning A Behavior, 1979,7,221-223.
W. A. Spatial memory
and radial arm maze performance.
Learning and Motivation,
G. E., &
memory and food searching strategies. In A. C. Kamil
& T. D.
(Eds.), Foraging behavior: Ecological, ethological, and
New York: Garland Press, 1981.
D. S., &
Remembrance of places
passed: Spatial memory in rats.
Behavior Processes, 1976,2,97-116.
F. H., &
Choice behavior of rats searching for food. Learning and Motiva-
1977, 8, 315-331.
W. A., & DALE, R. H. Remembrance of places lasts:
Proactive interference in rat spatial memory.
H. L. The meaning of representation in animal memory.
The Behavioral and Brain Sciences, in press. (a)
H. L. Decision making, evolution and cognition. In
H. D. Schmidt
& G. Tembrock (Eds.), Evolution and the deter-
animal and human behavior. New York: Elsevier-
North Holland, in press. (b)
H. M. The fighting fish of Siam. Natural History, 1958,
J. N. M. The food searching behaviour of two European
Description and analysis of search paths. Behaviour,
N. M. The food searching behaviour of two European
thrushes, 2. The adaptiveness
the search patterns. Behaviour,
1974,49, 1-61. (b)
H. H. Effects
experience on exploratory behavior in the rat.
T. G. Coevolution of foraging in Bombus and nectar
Chilopsis. A last dreg theory. Science, 1977, 197,
B. Ovenbirds (Aves: Perulidaei hunting
behaviour in a patchy environment: An experimental study.
Zoology, 1976,54, 1863-1879.
(Manuscript received June 7,1981;
revision accepted for publication September 25, 1981.)