Single motor unit and gross surface electromyographic responses to torque motor-produced wrist extensions were studied in human flexor carpi radialis muscle. Surface EMG typically showed two "periods" of reflex activity, at a short and long latency following stretch, but both periods occurring before a subject's voluntary reaction to the stretch. The amplitude of EMG activity in both reflex periods increased monotonically with an increase in the torque load. The amplitude of the short-latency reflex response was very dependent on the motoneuron pool excitability, or preload. The amplitude of the long-latency reflex response also varied with the preload, but could, in addition, be modulated by the subject's preparatory set for a voluntary response to the imposed displacement. When a single motor unit that was not tonically active began to fire during the stretch reflex, it did so primarily during the long-latency period. When caused to fire repetitively by voluntary facilitation of the motoneuron pool, that same unit now showed activity during both periods of the stretch reflex. Further increases in either motoneuron pool facilitation or in perturbation strength resulted in a monotonic increase in response probability of a single motor unit during the short-latency period. However, the response probability of a single unit during the long-latency reflex period did not always vary in a monotonic way with increases in either torque load or motoneuron pool facilitation. For an additional series of experiments, the subject was instructed on how to respond voluntarily to the upcoming wrist perturbation. The three instructions to the subject had no effect on the response probability of a single motor unit during either the background or short-latency periods of the stretch reflex. However, prior instruction clearly affected a unit's response probability during the long-latency reflex period. Changes in the firing rate of motor units, and in the recruitment or derecruitment of nontonic units, contributed to this modulation of reflex activity during the long-latency period.