1. The servo-like properties of muscle in healthy human subjects have been studied by interfering unexpectedly with flexion movements of the top joint of the thumb. This movement is carried out by the flexor pollicis longus muscle only. 2. The movements were standardized in rate by giving the subject a tracking task. They started off against a constant torque load offered by an electric motor. 3. In some movements the load remained constant, but in others, in mid-course, perturbations were introduced at random. Either the movement was halted, or released and allowed to accelerate by reducing the load, or reversed by suddenly increasing the current in the motor, so stretching the muscle. 4. Usually eight or sixteen responses to each kind of perturbation and a similar number of controls against a constant load were averaged. 5. Muscle activity was recorded as the electromyogram from surface electrodes over the belly of the long flexor in the lower forearm. Action potentials were usually full-wave rectified and integrated. 6. About 50 msec after a perturbation the muscle's activity alters in such a sense as to tend to compensate for the perturbation, i.e. it increases after a halt or a stretch and decreases after a release. The latency is similar in each case. 7. These responses are interpreted as manifestations of automatic servo action based on the stretch reflex. They are considered to be too early to be voluntary. 8. This interpretation was supported by measuring voluntary reaction times to perturbations under tracking conditions. They were found to be 90 msec or longer. 9. When the initial load was increased by a factor of 10, the servo responses were all scaled up likewise. Thus to a first approximation the gain of the servo is proportional to initial load. 10. It follows that in relaxed muscle the gain should be zero. This was confirmed by showing that stretching a relaxed muscle gives no reflex, or only a small one. 11. Gain appears to be determined by the level of muscle activation as determined by the effort made by the subject, rather than by the actual pressure exerted by the thumb. 12. Thus in fatigued muscle gain is boosted as the muscle has to be activated more strongly to keep up the same force output. The net effect is to compensate for fatigue and maintain the performance of the servo. 13. The Discussion centres on the implications of gain control in the servo. For a start, if the gain of the stretch reflex arc is zero in relaxed muscle, contractions cannot be initiated via the stretch reflex by simply causing the spindles to contract, as proposed on the original 'follow-up' servo theory.