Attentional bias toward negative social cues is thought to serve an etiological and/or maintaining role in social anxiety disorder (SAD). The current study tested whether training patients to disengage from negative social cues may ameliorate social anxiety in patients (N = 36) with a primary diagnosis of generalized SAD. Patients were randomly assigned to either an attention training condition (n = 18), in which patients completed a modified dot-probe task designed to facilitate attentional disengagement from disgusted faces, or a control dot-probe task condition (n = 18). As predicted, patients in the attention training condition exhibited significantly greater reductions in social anxiety and trait anxiety, compared with patients in the control condition. At termination, 72% of patients in the active treatment condition, relative to 11% of patients in the control condition, no longer met Diagnostic and Statistical Manual of Mental Disorders (4th ed.) criteria for SAD. At 4-month follow-up, patients in the attention training condition continued to maintain their clinical improvement, and diagnostic differences across conditions were also maintained. Results support attention-based models of anxiety and suggest that attention training is a promising alternative or complementary intervention.