Interpretation bias, defined as the tendency to interpret ambiguous stimuli as threatening, is thought to play a central role in the maintenance of social phobia. However, the effect of cognitive-behavioral therapy (CBT) on this bias has not been determined. The present study examined interpretation bias in individuals with generalized social phobia (GSP) who had been treated with CBT, untreated GSPs, and nonanxious controls (NACs). Results indicated that untreated GSPs had higher negative bias scores than treated GSPs and NACs on a forced-choice measure of interpretation bias, although the latter two groups did not differ. Following study hypotheses, the interpretation bias found in untreated GSPs was specific to social items, and was evident when raters coded participants' initial written responses to the scenarios, when compared with control subjects. Treated patients had open-ended responses that did not differ from controls or untreated GSPs. Findings are discussed in the context of other GSP interpretation bias studies, and future research directions are recommended.