Safety behavior (SB) and self-focused attention (SFA) have been posited as important maintenance factors in the cognitive model of social anxiety disorder (SAD). The present study reports the results of experiments to drop SB and SFA among clinically diagnosed patients with SAD employing their own idiosyncratic anxiety-provoking situations. The ratings for observable anxiety, belief in feared outcome and overall performance were better for role plays without SB and SFA than for role plays with them. The degree of drop in SFA predicted drop in observable anxiety and belief in feared outcome. Dropping SB and SFA, however, was unable to completely correct the cognitive distortion because the subjective ratings were still significantly worse than the objective ratings.