Forty-three college students suffering from recurrent tension headache were randomly assigned to 1 of 4 electromyographic (EMG) biofeedback training conditions. Although all subjects were led to believe they were learning to decrease frontal EMG activity, actual feedback was contingent on decreased EMG activity for half of the subjects and increased EMG activity for the other half. Within these 2 groups, subjects also viewed bogus video displays designed to convince them they were achieving large (high success) or small (moderate success) reductions in EMG activity. Regardless of actual changes in EMG activity, subjects receiving high-success feedback showed substantially greater improvement in headache activity (53%) than subjects receiving moderate success feedback (26%). Performance feedback was also related to changes in locus of control and self-efficacy. Changes in these 2 cognitive variables during biofeedback training were also correlated with reductions in headache activity following treatment, whereas changes in EMG activity exhibited during training were uncorrelated with outcome. These results suggest that the effectiveness of EMG biofeedback training with tension headache may be mediated by cognitive changes induced by performance feedback and not primarily by reductions in EMG activity.