Interpreting Neutral Faces as Threatening Is a Default Mode for Socially
K. Lira Yoon and Richard E. Zinbarg
The authors of the present study used an incidental learning paradigm to investigate the interpretation of
neutral facial expressions in socially anxious individuals. Participants were asked to detect the location
of a target following the presentation of a facial picture (i.e., cue). Unbeknownst to participants, the target
location was contingent on the valence of the cue, and participants thus learned to associate different
target locations with either positive or negative facial expressions. The authors subsequently used this
learned association to assess interpretive biases. If socially anxious individuals interpret neutral faces in
a negative manner, they should be faster to detect a target that appears in the location that is associated
with negative face cues when the target is presented after a neutral face cue. The authors also assessed
whether the anticipation of a feared situation influenced interpretive biases by comparing participants
with and without a speech threat on this task. Results indicate that socially anxious individuals are
characterized by an interpretive bias regardless of the threat manipulation. In contrast, nonanxious
individuals interpreted neutral faces in a negative manner only when they were in the threat condition.
Keywords: social anxiety; interpretive bias; face
As the basic fear in social anxiety (SA) is that of receiving
negative evaluation, which is often conveyed by and inferred
from facial expressions, interpretations of ambiguous facial
expressions might be particularly relevant for understanding SA
(Philippot & Douilliez, 2005). Evidence for a facial interpretive
bias in socially anxious individuals, however, is weak at best.
Although a few studies have reported that socially anxious
individuals interpret neutral faces in a negative manner (e.g.,
Lundh & O
¨st, 1996a), others have failed to do so (e.g., Lundh
¨st, 1996b). Even more problematic, the studies reporting
significant findings are mostly limited by their sole reliance on
self-reports. It is, thus, possible that the results from previous
studies might reflect socially anxious individuals’ tendency to
select negative response options (i.e., response bias).
We employed incidental learning methodology (e.g., Buchner
& Wippich, 1998; Lewicki, 1986) to compare high-SA individ-
uals with low-SA individuals in their interpretations of pictures
of ambiguous facial expressions. As participants were not asked
to direct their attention to the valence of the stimuli or to engage
in conscious categorization of stimuli in terms of valence, we
were able to study interpretive bias that could not be readily
explained by response bias or demand.
The participants’ task was to locate a target and then to press
a button corresponding to its location as soon as possible.
Unbeknownst to the participants, there was a relationship be-
tween the valence of a face stimulus (i.e., cue) and the target
location that participants were exposed to during the learning
phase. In the learning phase, negative and positive faces served
as cues. If learning occurred, participants should more rapidly
detect the target appearing at the expected location according to
the pattern presented in the learning phase. For the critical trials
in the testing phase, neutral faces—inherently ambiguous as to
the emotional state of the actor—appeared as cues.
It is important to note that threat manipulations often interact
with individual differences in trait anxiety in regard to cognitive
biases (e.g., Mathews & MacLeod, 1994). The results are
inconsistent with some studies that have suggested threat ma-
nipulations enhance biases (e.g., Calvo & Castillo, 1997;
Mansell, Clark, & Ehlers, 2003), whereas other studies have
suggested that threat manipulations suppress biases in the anx-
ious populations (e.g., Amir et al., 1996; Mathews & Sebastian,
1993). It thus seems important to examine the effects of threat
manipulation on interpretive bias. Therefore, half the partici-
pants were assigned to a speech condition, and half were
assigned to a no-speech condition.
On the basis of cognitive models of social phobia (Clark &
Wells, 1995; Rapee & Heimberg, 1997), we hypothesized that
socially anxious individuals would interpret neutral faces in a
negative manner (i.e., they would show relative facilitation in
target detection when the target appears at the location that they
learned to associate with the negative faces following a neutral
K. Lira Yoon, Department of Psychology, Northwestern University;
Richard E. Zinbarg, Department of Psychology, Northwestern University;
and The Family Institute, Northwestern University.
Correspondence concerning this article should be addressed to K. Lira
Yoon, who is now at the Department of Psychology, University of Miami,
P.O. Box 248185, Coral Gables, FL 33124-0751. E-mail:
Journal of Abnormal Psychology Copyright 2008 by the American Psychological Association
2008, Vol. 117, No. 3, 680– 685 0021-843X/08/$12.00 DOI: 10.1037/0021-843X.117.3.680
One hundred sixteen introductory psychology students com-
pleted the Social Phobia Scale (SPS; Mattick & Clarke, 1998)
during an initial group testing session. Sixty students who scored
in the top (SPS ⱖ18) or bottom (SPS ⱕ7) 25th percentile on the
SPS were then chosen to participate in the study as part of a course
requirement. Due to technical failure, 11 participants were unable
to complete the incidental learning task. In addition, 2 participants
had at least one missing cell and thus were excluded from analyses.
The loss of data was spread evenly across SA level (i.e., 7
participants in the low-SA group and 6 in the high-SA group). The
final sample consisted of 47 participants (35 female), with 23 (17
female) low-SA (SPS, M⫽5.91, SD ⫽3.73) and 24 (18 female)
high-SA participants (SPS, M⫽28.75, SD ⫽8.36). About two
thirds of our high-SA participants met or exceeded the score of 26
suggested by Peters (2000) to identify those with social anxiety
Symptom measures. The SPS assesses a participant’s typical
levels of fear of scrutiny when performing a task or being observed
by others and has adequate reliability and concurrent validity.
Participants also completed the Beck Depression Inventory (BDI;
Beck, Ward, Mendelson, Mock, & Erbaugh, 1961), which has
adequate reliability and has been extensively validated (e.g., Beck,
Steer, & Garbin, 1988). Alpha in the current study equaled .93 for
the SPS and .87 for the BDI. Participants’ state anxiety level was
measured by a single item (“How anxious do you feel right now?”)
ona0(not at all)to9(extremely) scale on the computer.
Face stimuli. Angry, disgusted, happy, and neutral faces were
obtained from Ekman and Friesen’s (1976) Pictures of Facial
Affect. Fourteen models (8 female, 6 male) posed for each facial
affect, resulting in a total of 56 different facial stimuli. Eight
neutral pictures appeared once, and the other 6 appeared twice; 6
happy pictures appeared eight times, and the other 8 appeared nine
times; 10 angry pictures and 10 pictures showing disgust appeared
four times, and the remaining 4 angry and 4 disgusted pictures
appeared five times.
Incidental Learning Task
The task consisted of a learning phase and a testing phase. Each
trial consisted of a 500-ms fixation point, a facial picture (cue)
presented for 675 ms that was followed immediately by a target
(letter L), and a question that asked the participant whether the
person in the picture was a man or a woman to ensure attention to
During the learning phase, the cue was either negative (i.e., an
angry or a disgusted face) or positive (i.e., a happy face). Once the
cue disappeared, the target appeared at any of four premarked
positions on the computer screen (see Figure 1). When the cue was
negative, the target appeared at either the top or the bottom two
locations (i.e., negative location) on 80% of the trials. The side
opposite to the negative location was the positive location. For
example, for some participants when the cue was negative, the
target appeared at the upper-left location 40% of the time, at the
upper-right location 40% of the time, and at each of the two lower
locations 10% of the time. For these participants, when the cue was
positive, the target appeared at the bottom-left 40% of the time and
at the bottom-right 40% of the time. Whether the top or the bottom
was designated as the negative location was determined randomly
for each of the participants. For the final sample, there was no
difference in the number of participants from each group who
received the top as the negative location.
Participants were instructed to respond to the appearance of the
target as quickly as possible by pressing one of four keys that
corresponded to the locations of the target. We measured response
latency as the time interval between the onset of the target display
and the response. Each participant was presented with 10 blocks of
learning trials, with each block consisting of 20 trials (i.e., 10 trials
with negative cues—five angry and five disgusted—and 10 trials
with positive cues) in a different randomized order.
During the testing phase, neutral face cues appeared in addition
to positive and negative face cues. The relationship between the
valence of the cue and the target location during the testing phase
remained identical for the positive and the negative cues. For the
neutral cue trials, the target appeared at any of the four possible
positions with equal chances. There were 20 trials each with
neutral, positive, and negative cues. We used the same positive and
negative face cues as we did in the learning phase.
Participants were tested in groups of three to four. After signing
the consent form, participants in the speech condition were asked
to prepare for a speech for 2 min, whereas participants in the
no-speech condition were asked to write an essay. Participants in
the no-speech condition were assured that their essays would not
a) b) c) d)
Figure 1. Illustration of the sequence of events on a single trial. The trial began with a fixation point (a), which
stayed on the screen for 500 ms. The fixation point was then replaced with a face stimulus (i.e., cue). During
the learning phase, the cue was either a negative or a positive expression. During the testing phase, a negative,
positive, or neutral cue appeared. The target (Letter L) appeared at any of four possible locations and stayed on
the screen until the participant made a response (c). Finally, a question appeared on the screen asking the
participant whether the picture was of a man or a woman (d).
be read by anyone but that the experimenter would check whether
they wrote an essay. Participants in the speech condition were told
that one of them would be randomly selected to give a videotaped
speech toward the end of the session. Next, participants were asked
to complete the computer task. Participants were told that their task
was to identify the location of the Lby pressing one of the four
keys that corresponded to the location of the target. After respond-
ing to the target, participants indicated whether the person in the
picture was a man or a woman.
Participants began the incidental learning task by completing the
single-item state anxiety measure. One of the participants in each
speech group was asked to give a 2-min videotaped speech once
everyone in the group completed the incidental learning task. After
the speech (or after the computer task, if they were in the no-
speech group), participants were given a questionnaire that as-
sessed awareness of the contingencies.
A series of SA ⫻Speech analyses on age, gender, BDI, and SPS
scores were conducted. Socially anxious individuals scored higher
on the BDI, F(1, 43) ⫽12.16, p⬍.001, and on the SPS, F(1,
43) ⫽139.46, p⬍.001, than their counterparts. No other effects
were significant, indicating that randomization was successful.
To test whether the speech manipulation had its intended effect,
we examined participants’ anxiety levels right before they began
the computer task (i.e., after they learned their experimental con-
dition). The SA ⫻Speech analysis of variance (ANOVA)
vealed significant main effects for SA, F(1, 43) ⫽12.08, p⬍.001,
⫽.05, and for speech, F(1, 43) ⫽4.30, p⬍.05,
pattern confirmed that groups differed in the manner that we
intended by the manipulation (speech, M⫽3.83, SD ⫽1.81;
no-speech, M⫽2.87, SD ⫽1.87). Importantly, the interaction
between SA and speech was not significant, F(1, 43) ⫽2.06,
.007. Means and standard deviations of state anxiety level for each
group was as follows: low-SA, no-speech (M⫽2.36, SD ⫽1.86);
low-SA, speech (M⫽2.67, SD ⫽1.30); high-SA, no-speech (M⫽
3.33, SD ⫽1.83); high-SA, speech (M⫽5.00, SD ⫽1.48).
Data from trials with errors (i.e., incorrect identification of the
target location or incorrect identification of the gender of a cue)
were discarded. In addition, response times (RTs) that were less
than 200 ms or greater than 1,500 ms were excluded on the basis
of the boxplot (cf. Barnett & Lewis, 1994). For the final sample,
the mean percentages of data lost due to errors and outliers were
2.9% and 1.5%, respectively. Errors and outliers did not differ by
group, speech condition, or cue type (all Fs⬍1.1). The overall
mean RT was 609.1 ms (SD ⫽451.4) before and 563.9 ms (SD ⫽
122.5) after the data cleanup. Because our RT data were not
normally distributed (Shapiro–Wilk ⫽.89, p⬍.001), we reana-
lyzed the data with log-transformed RTs, which successfully nor-
malized the data (Shapiro–Wilk ⫽.96, ns). The results remained
the same, therefore we report the analyses based on raw RTs for
the simplicity of interpretation of results.
Learning of the Cue Valence and Target Location
To examine whether participants learned the relationship be-
tween cue valence and the target location and to rule out the
possibility that socially anxious individuals learned the association
involving the negative cues more strongly than did their less
anxious counterparts, we carried out an SA (high vs. low) ⫻Cue
Type (negative vs. positive) ⫻Target Location (expected vs.
unexpected) ⫻Block mixed ANOVA with polynomial trend anal-
ysis on the RTs for the training trials.
As expected, the Target
Location ⫻Linear Trend of Block interaction was significant, F(1,
45) ⫽4.53, p⬍.04,
⫽.01; the decrease in RT was steeper over
blocks for the expected target location (slope ⫽⫺.007) compared
with the unexpected target location (slope ⫽⫺.003). The Cue
Type ⫻Target Location ⫻Linear Trend of Block interaction was
not significant (F⬍1). Importantly, the four-way interaction,
which might indicate differential acquisition of the associations
between the groups, was not significant (F⬍1). In addition,
neither the SA ⫻Target Location ⫻Block interaction nor the
SA ⫻Target Location was significant (both Fs⬍1). Therefore,
these results provide no support for a stronger expectation for the
threat location among the socially anxious participants.
Interpretive Bias and SA
Although the results above suggest that participants learned the
contingencies to the same extent regardless of their SA level, it is
still possible for socially anxious individuals to become faster at
detecting the target that appears at the negative location regardless
of the valence of the cue on the testing trials. Similarly, partici-
pants in the speech condition might have become better at detect-
ing the target that appeared at the negative location due to their
heightened levels of anxiety. To rule out these possibilities and to
test the main hypothesis that socially anxious individuals interpret
neutral faces negatively, a SA ⫻Speech ⫻Cue Type ⫻Target
Location ANOVA was conducted on the RTs during the testing
phase. There was a significant main effect of target location, F(1,
43) ⫽7.65, p⬍.01,
⫽.15, which was qualified by a significant
four-way interaction, F(2, 42) ⫽3.69, p⬍.04,
interaction between SA and target location, however, was not
significant, suggesting that socially anxious individuals were not
significantly different from their less-anxious counterparts in their
readiness to detect targets appearing at locations associated with
negative cues. Likewise, the participants in the speech condition
were not significantly different from those in the no-speech con-
dition in their tendency to detect targets that appear at locations
associated with negative cues as indicated by a nonsignificant
Speech ⫻Target Location interaction.
Because the size of a group that a participant was in could affect the
level of state anxiety, an additional ANOVA with SA, speech condition,
and group size was conducted. The results remained the same, and none of
the effects involving the group size was significant.
Mean RTs for each block are available upon request from K. Lira
682 BRIEF REPORTS
To follow up on the significant four-way interaction, we tested
the simple three-way interaction between SA, speech, and target
location separately for neutral, positive, and negative cues. The
simple three-way interaction was significant for neutral cues, F(1,
43) ⫽10.56, p⬍.01,
⫽.17, but not for positive or negative
cues (both Fs⬍1), indicating that the group differences in
responses to the neutral cues were responsible for the four-way
interaction. Table 1 presents the means in each condition.
To break down the simple three-way interaction for neutral cues,
we calculated a speeding index by subtracting the RT when the
target appeared at the positive location from the RT when the
target appeared at the negative location. Thus, negative speeding
index scores reflect a tendency to respond more rapidly when the
target appeared at the negative location, which is consistent with
interpreting neutral faces in a more negative way. Simple effects
analyses revealed that the high-SA group showed significantly
more negative bias than the low-SA group in the no-speech con-
dition, F(1, 21) ⫽8.16, p⬍.01,
⫽.28, but not in the speech
condition, F(1, 22) ⫽2.20, ns. In addition, the effect of speech
condition was significant for the low-SA group, F(1, 21) ⫽9.81,
⫽.32, but not for the high-SA group, F(1, 22) ⫽2.69,
ns. The mean speeding index for the speech condition, collapsed
across the anxiety levels, was significantly different from 0,
t(23) ⫽⫺2.66, p⬍.01.
For those in the no-speech condition, the
speeding index of high-SA participants was significantly different
from 0, t(11) ⫽⫺2.26, p⬍.05, whereas the speeding index of
low-SA participants was not, t(10) ⫽1.82, ns. These results
suggest that everyone exhibited negative interpretive bias when
they were in the speech condition (see Figure 2). However, only
the high-SA individuals exhibited negative interpretive bias in the
To further provide anchors against which to determine whether
observed patterns signify negative or positive interpretive bias, we
computed speeding indexes for negative and positive cues in the
same manner as we created the speeding index for the neutral cue.
That is, the RTs to detect targets that appeared at positive locations
were subtracted from the RTs of detected targets that appeared at
negative locations. When the speeding indexes for the neutral cues
were statistically contrasted to the speeding indexes for the unam-
biguous cues, the speeding index for the neutral cues was signif-
icantly lower than it was for positive cues in the no-speech,
high-SA group, t(11) ⫽2.78, p⬍.05. No other comparisons were
Through our use of an incidental learning paradigm, the results
suggest an interpretive bias in socially anxious individuals that
cannot be readily explained by either response bias or experi-
menter demand. Thus, the present results suggest that highly
socially anxious individuals treat neutral social interaction cues as
conveying anger or disgust and/or contempt, thereby extending the
results of earlier studies in which researchers used verbal stimuli
(e.g., Hirsch & Mathews, 2000). These findings also replicate and
extend results obtained in a previous study in which researchers
used a priming paradigm (Yoon & Zinbarg, 2007). Although
demand characteristics cannot be conclusively refuted, it seems
implausible that the participants would have figured out (a) our
hypothesis that SA would be related to interpretive bias and (b)
how this hypothesis translated into predictions about their RTs.
Another purpose of the current study was to examine whether
the interpretive bias would be affected by a threat manipulation. In
the current study, the group differences were significantly smaller
when there was an explicit threat (i.e., speech condition). In fact,
the group differences were only significant when there was no
apparent threat, such that high-SA individuals exhibited a negative
interpretive bias, whereas the low-SA individuals did not exhibit
such a bias. In contrast, participants, regardless of anxiety levels,
exhibited negative interpretive biases in the threat condition. Thus,
it could be argued that interpreting neutral faces in a negative
manner is the default mode for high-SA individuals, whereas
low-SA individuals interpret neutral faces negatively only when
One explanation for the effect of the speech threat draws on the
model proposed by Mathews, Mackintosh, and Fulcher (1997).
Because there was no significant group difference in the speech con-
dition, we tested only one mean collapsed across the anxiety levels to
Mean Response Latencies to Targets (in ms) for Neutral, Negative, and Positive Cue Trials
Low-SA group High-SA group
Speech (n⫽12) No-speech (n⫽11) Speech (n⫽12) No-speech (n⫽12)
Negative 467.71 57.77 593.16 201.21 509.85 107.51 527.44 73.14
Positive 514.13 96.38 553.55 169.24 523.26 103.74 594.43 131.66
Negative 501.66 140.77 585.64 186.26 520.87 74.55 557.95 111.73
Positive 557.49 143.85 636.69 297.85 524.14 81.58 660.25 169.65
Negative 549.17 240.51 625.68 212.66 546.73 97.20 644.25 135.04
Positive 507.23 95.60 594.61 163.71 517.28 142.94 580.06 99.11
According to Mathews et al., interpretive biases arise from selec-
tive attention to threat, which enhances the activation of the
threatening interpretation of an ambiguous stimulus. More specif-
ically, we assume that when faced with the threat of public
speaking, most people experience enhanced selective attention to
potentially threatening faces. For the low-SA group, this resulted
in a boost of the activation of the negative interpretation of an
ambiguous face such that it tended to win the mutually inhibitory
competition among the activated interpretations. For the high-SA
group, the activation of the negative interpretation was already
boosted to the point at which the negative interpretation was
already tending to win even in the absence of the speech threat.
What remains unclear are the conditions in which a threat
manipulation (a) enhances group differences in bias (e.g., Calvo &
Castillo, 1997), (b) masks group differences by causing the
low-SA group to demonstrate the bias otherwise characteristic of
the high-SA group (as presented here), or (c) masks group differ-
ences by suppressing the bias otherwise demonstrated by the
high-SA group (e.g., Mathews & Sebastian, 1993). Studies differ
in several ways, including the type of bias assessed (i.e., atten-
tional vs. interpretive), the type of stimuli presented (e.g., verbal,
pictorial), the nature of the populations studied (i.e., clinical vs.
high-trait anxious), and the temporal location of a threat. Further
research needs to be done to sort out which methodological dif-
ferences account for the differences in the results.
Limitations of this study are worth mentioning. First, we did not
administer diagnostic interviews. Thus, future studies are neces-
sary to test whether the obtained results can be generalized to a
clinical population. Second, we do not have information about
whether participants in the speech condition believed that they had
to give a speech. However, considering the higher levels of anxiety
in the speech condition compared with the no-speech condition,
the speech manipulation seemed to have its intended effect. It is
also possible that our low-SA group might have included individ-
uals with high public speaking anxiety who are otherwise not
socially anxious. This could have contributed to the findings of
low-anxiety individuals in the speech condition who showed neg-
ative biases. Third, we used a single-item measure to assess the
level of state anxiety. The item has good face validity and dem-
onstrated that the speech manipulation had its intended effects on
the level of state anxiety. It is still possible, however, that the
single item might not have adequately assessed the group differ-
ences due to the threat manipulation. Relatedly, our results might
represent a levels effect; that is, an increase from low to moderate
state anxiety may serve to elicit a negative interpretive bias, which
is unaffected by further increases in state anxiety. Despite the
nonsignificant interaction between the group and the speech con-
dition, the pattern of means suggests that our threat manipulation
moved individuals with different levels of SA through different
parts of the state anxiety continuum. Thus, it remains for future
researchers to determine whether the difference between the anx-
iety groups that are not undergoing a social challenge in interpre-
tive response is due to the inherent state anxiety difference be-
tween the groups or to a different impact of state anxiety on
interpretive bias in the two groups.
One might argue that neutral faces are not ambiguous. We note
that one can draw different inferences about the actor’s emotional
state from a neutral expression, because there is considerable
ambiguity in this process when the actor displays a neutral expres-
sion. Relatedly, Ekman and Friesen’s (1976) norms indicate that
the neutral faces tend to be more ambiguous than other facial
expressions, as there is far more variance in responses to the
Low SA High SA
Figure 2. Bias score for neutral cue trials as a function of social anxiety (SA) and speech condition. Error bars
represent one standard error.
684 BRIEF REPORTS
neutral faces. It is possible that the present results might reflect
socially anxious individuals’ negative emotional reactions to faces
that they and the low-SA group interpret identically. Differential
emotional appraisal alone, however, cannot explain the whole
story. If the groups are making different emotional appraisals of
the neutral faces, and socially anxious individuals find all faces
more aversive than their less-anxious counterparts, we are still left
with a continuum of relative aversiveness within each group. That
is, differential emotional appraisal cannot fully account for why
the socially anxious individuals tend to respond to neutral faces
more as they do the truly aversive faces than they do the happy
faces. Another important extension for future studies would be to
use nonsocial stimuli in the learning phase to even more defini-
tively rule out that the current results are due to group differences
in initially learning about the positive and negative social stimuli.
These limitations notwithstanding, the incidental learning para-
digm renders response bias implausible, as participants were never
asked to indicate what expression they thought was represented in
each of the pictures. Thus, the present study provides strong
support for the hypothesis that SA is associated with a bias toward
interpreting neutral facial expressions in a negative manner. More
specifically, our results suggest that interpreting neutral faces in a
negative manner is the default mode for socially anxious individ-
uals, whereas people with low levels of anxiety interpret neutral
faces negatively only when they are under threat.
Amir, N., McNally, R. J., Riemann, B. C., Burns, J., Lorenz, M., & Mullen,
J. T. (1996). Suppression of the emotional Stroop effect by increased
anxiety in patients with social phobia. Behaviour Research and Therapy,
Barnett, V., & Lewis, T. (1994). Outliers in statistical data. New York:
Beck, A. T., Steer, R. A., & Garbin, M. G. (1988). Psychometric properties
of the Beck Depression Inventory: Twenty-five years of evaluation.
Clinical Psychology Review, 8, 77–100.
Beck, A. T., Ward, C. H., Mendelson, M., Mock, J., & Erbaugh, J. (1961).
An inventory for measuring depression. Archives of General Psychiatry,
Buchner, A., & Wippich, W. (1998). Differences and commonalities be-
tween implicit learning and implicit memory. In M. A. Stadler & P. A.
Frensch (Eds.), Handbook of implicit learning (pp. 3– 46). Thousand
Oaks, CA: Sage.
Calvo, M. G., & Castillo, M. D. (1997). Mood-congruent bias in interpre-
tation of ambiguity: Strategic processes and temporary activation. Quar-
terly Journal of Experimental Psychology: Human Experimental Psy-
chology, 50A, 163–182.
Clark, D. M., & Wells, A. (1995). A cognitive model of social phobia. In
R. G. Heimberg, M. R. Liebowitz, D. A. Hope, & F. R. Schneier (Eds.),
Social phobia: Diagnosis, assessment, and treatment (pp. 69 –93). New
York: Guilford Press.
Ekman, P., & Friesen, W. V. (1976). Pictures of facial affect [Brochure].
Palo Alto, CA: Consulting Psychologists Press.
Hirsch, C. R., & Mathews, A. (2000). Impaired positive inferential bias in
social phobia. Journal of Abnormal Psychology, 109, 705–712.
Lewicki, P. (1986). Processing information about covariations that cannot
be articulated. Journal of Experimental Psychology: Learning, Memory,
and Cognition, 12, 135–146.
Lundh, L. G., & O
¨st, L. G. (1996a). Face recognition in patients with social
phobia. Scandinavian Journal of Behaviour Therapy, 25, 139 –148.
Lundh, L. G., & O
¨st, L. G. (1996b). Recognition bias for critical faces in
social phobics. Behaviour Research and Therapy, 34, 787–794.
Mansell, W., Clark, D. M., & Ehlers, A. (2003). Internal versus external
attention in social anxiety: An investigation using a novel paradigm.
Behaviour Research and Therapy, 41, 555–572.
Mathews, A., Mackintosh, B., & Fulcher, E. P. (1997). Cognitive biases in
anxiety and attention to threat. Trends in Cognitive Science, 1, 340 –345.
Mathews, A., & MacLeod, C. (1994). Cognitive approaches to emotion and
emotional disorders. Annual Review of Psychology, 45, 25–50.
Mathews, A., & Sebastian, S. (1993). Suppression of emotional Stroop
effects by fear-arousal. Cognition and Emotion, 7, 517–530.
Mattick, R. P., & Clarke, J. C. (1998). Development and validation of
measures of social phobia scrutiny fear and social interaction anxiety.
Behaviour Research and Therapy, 36, 455– 470.
Peters, L. (2000). Discriminant validity of the Social Phobia and Anxiety
Inventory (SPAI), the Social Phobia Scale (SPS) and the Social Inter-
action Anxiety Scale (SIAS). Behaviour Research and Therapy, 38,
Philippot, P., & Douilliez, C. (2005). Social phobics do not misinterpret
facial expression of emotion. Behaviour Research and Therapy, 43,
639 – 652.
Rapee, R. M., & Heimberg, R. G. (1997). A cognitive-behavioral model of
anxiety in social phobia. Behaviour Research and Therapy, 35, 741–756.
Yoon, K. L., & Zinbarg, R. E. (2007). Threat is in the eye of the beholder:
Social anxiety and the interpretation of ambiguous facial expressions.
Behaviour Research and Therapy, 45, 839 – 847.
Received December 11, 2006
Revision received January 18, 2008
Accepted March 3, 2008 䡲