Successful treatment of torticollis with electromyographic (EMG) biofeedback has been reported in a number of single case and single group studies. The present investigation represents the first controlled outcome study. Twelve torticollis patients were randomly assigned to EMG biofeedback or relaxation training and graded neck exercises (RGP). The procedure involved three sessions of baseline assessment, 15 sessions of EMG BF or RGP, 6 sessions of EMG BF or RGP plus home-management, 6 sessions of home-management alone, and follow-up 3 months after the end of treatment. A variety of outcome measures were used including physiological (EMG from the two sternocleidomastoid muscles, skin conductance level), behavioral (angle of head deviation, range of movement of the head), and self-report (depression, functional disability, body concept), therapist and "significant other" reports and independent observer assessment of videos. In both groups, neck muscle activity was reduced from pre- to posttreatment. This reduction was greater in the EMG biofeedback group. There was evidence of feedback-specific neck muscle relaxation in the EMG biofeedback group. Therefore, the outcome was not due to nonspecific factors and could be attributed to feedback-specific effects. Changes in skin conductance level showed that neck muscle relaxation was not simply mediated by a general reduction of "arousal." Significant improvements of extent of head deviation, and range of movement of the head, as well as reductions of depression were present, which were not different in the two groups. At the end of treatment, no patient was asymptomatic. Any therapeutic benefit was generally maintained at follow-up. The results and the procedural simplicity of RGP make the issue of cost-efficacy of EMG biofeedback a pertinent one. Further controlled outcome studies of EMG biofeedback treatment of torticollis with larger samples are required.