The present study examined the reciprocal relationships between probability estimates, cost estimates, and social anxiety during a 12-week course of cognitive-behaviour therapy (CBT) for social anxiety disorder (SAD). One hundred and sixty-three individuals with a principal diagnosis of SAD completed a weekly tracking measure that included questions about probability and cost estimates for negative consequences and current levels of social anxiety relating to a hypothetical speech task. Paired-samples t tests revealed that these variables changed from pre-to-post treatment. Cross-lagged path analyses revealed that cost estimates predicted subsequent changes in social anxiety levels early in treatment, over and above previous anxiety ratings, while probability estimates predicted subsequent levels of social anxiety at the final stages of treatment, over and above previous anxiety ratings. Overall, the results indicated that (1) cost bias early in treatment predicts subsequent social anxiety levels, and (2) both probability and cost estimates seem to play important, but seemingly independent, roles as mechanisms of change in CBT for SAD.