ObjectiveThe present investigation examined (a) whether a clinical sample of individuals with social anxiety disorder (SAD) comprises two distinct groups based on attention bias for social threat (vigilant, avoidant), (b) the relation between attention bias and cognitive bias, specifically estimates of the probability that negative social events will occur (probability bias), and (c) specific changes in attention bias following cognitive behavioral therapy for social anxiety.MethodParticipants were 24 individuals (nfemale = 7, nmale = 17; mage = 41) who met diagnostic criteria for SAD and sought treatment for fear of public speaking. Hypotheses were tested using t tests, linear regression analyses, and a mixed design analysis of variance.ResultsResults yielded evidence of 2 pretreatment groups (vigilant and avoidant). There was a significant positive correlation between vigilance for (but not avoidance of) threat and probability bias (R = .561, p < .05). After 8 weeks of treatment, the direction of change in attention bias differed between groups, such that the vigilant group became less vigilant and the avoidant group became less avoidant, with the avoidant group showing a significant change in attention bias from pretreatment to posttreatment.ConclusionsThese findings provide very preliminary support for the idea that individuals with SAD may differ according to type attention bias, avoidant or vigilant, as these biases changed in different ways following cognitive‐behavioral therapy for SAD. Further research is needed to replicate and extend these findings in order to evaluate whether SAD comprises subgroups of attentional biases. © 2012 Wiley Periodicals, Inc. J. Clin. Psychol. 68:745‐754, 2012