Cognitive theories posit that exposure-based treatments exert their effect on social anxiety by modifying judgmental biases. The present study provides a conservative test of the relative roles of changes in judgmental biases in governing social anxiety reduction and addresses several limitations of previous research. Longitudinal, within-subjects analysis of data from 53 adults with a Diagnostic and Statistical Manual of Mental Disorders (4th ed.; American Psychiatric Association, 1994) social phobia diagnosis revealed that reductions in probability and cost biases accounted for significant variance in fear reduction achieved during treatment. However, whereas the reduction in probability bias resulted in fear reduction, the reduction in cost bias was merely a consequence of fear reduction. A potential implication is that exposure-based treatments for social anxiety might focus more attention on correcting faulty appraisals of social threat occurrence.