The aim of this study was to investigate the effect of mindfulness-based stress reduction (MBSR) for patients with heterogeneous anxiety disorders. Seventy-six self-referred patients were randomized to MBSR or a waiting-list control condition. Eight participants did not complete the eight-week MBSR intervention. Treatment completers improved significantly on all outcome measures compared to controls. The completer sample showed medium to large effect sizes on measures of anxiety (Cohen's d = 0.55-0.97), and a large effect size for symptoms of depression (Cohen's d = 0.97). Intention-to-treat analyses yielded effect sizes in the small to moderate range (Cohen's d = 0.32-0.76). Gains were maintained at six months follow-up. The percentage of participants reaching recovered status was highest for symptom measures of depression and anxiety, and lower for worry and trait anxiety. Mediation analyses indicated that mindfulness fully mediated changes in acute anxiety symptoms, and partially mediated changes in worry and trait anxiety. However, the present study did not find evidence of temporal precedence for the proposed mediator. In the absence of true mediation and an active control condition, it cannot be ruled out that results are due to non-specific aspects of treatment. Despite these and other limitations, we conclude that MBSR is an effective treatment for anxiety disorders and related symptomatology.