Voluntary actions produce suppression of neural activity in sensory areas, and reduced levels of conscious sensation. Recent computational models of motor control have linked sensory suppression to motor prediction: an efferent signal from motor areas may cancel the sensory reafferences predicted as a consequence of movement. Direct evidence for the efferent mechanism in sensory suppression has been lacking. We investigated the perceived size of finger-muscle twitches (MEPs) evoked by TMS in eight normal subjects. Subjects freely chose on each trial whether to make or withhold a voluntary flexion of the right index finger, in synchrony with an instructional stimulus. A test MEP occurred at the instructed time of action. The subject then relaxed and a second reference MEP occurred a few seconds later. Subjects judged which of the two MEPs was larger. Subjects perceived the first test MEP to be smaller in trials where they made voluntary actions than on trials where they did not, demonstrating sensory suppression. On randomly selected trials, a conditioning prepulse was delivered over the supplementary motor area (SMA) 10 ms before the pulse producing the test MEP. The SMA prepulse reduced and almost abolished the sensory suppression effect in voluntary action trials. We suggest the SMA may provide an efferent signal which is used by other brain areas to modulate somatosensory activity during self-generated movement.