Judgmental biases for threat-relevant stimuli are thought to be important mechanisms underlying the etiology and maintenance of anxiety disorders. The authors hypothesized (a) that people with generalized social phobia (GSP) would rate negative social events but not nonsocial events as more probable and costly than would nonanxious controls (NACs) and (b) that cognitive behavioral treatment would decrease probability and cost estimates for social but not nonsocial events. Participants with GSP and NACs were assessed twice, 14 weeks apart, during which the former received cognitive behavioral therapy. Those with GSP evidenced socially relevant judgmental biases prior to treatment, and these were attenuated following treatment. Reduction in cost estimates for social events, but not in probability estimates, mediated improvement in social phobia. Results are discussed in light of emotional processing theory.