We have investigated the role of reflex regulation of muscle force in normal human subjects by comparing changes in the stretch-evoked increments in elbow joint flexor electromyogram (EMG) and elbow joint torque before and after fatigue. Elbow flexor muscle fatigue was induced by repetitive voluntary isometric contractions. To assess the appropriateness of the EMG signal as an index of neural excitation of muscle under fatiguing conditions, we examined the time course of recovery of joint torque and EMG power spectrum following fatigue. Fatigue-related changes in the EMG power spectra recovered within 5-10 min after fatiguing exercise was terminated, yet the muscle weakness induced by the exercise lasted greater than 7 h and was substantial in magnitude. The decoupling of torque and EMG recovery allowed us to compare pre- and postfatigue EMG stretch responses without adjusting for differences in EMG spectral content. Torque and EMG responses to stretch were quantified by time-averaging over 250-ms "isometric" and "steady-state" periods, just before and just after a ramp angular stretch of the elbow joint, respectively. The torque increment elicited by stretch was lower following fatigue in seven of eight experiments. However, the average decrease of 20.13 +/- 14.42% in these seven subjects was somewhat smaller than the corresponding average shift in the slope of the isometric EMG-torque relationship of 85.84 +/- 90.29% (n = 8). Furthermore, the stretch-induced EMG increment was larger following fatigue in all eight sessions (average of 56.14 +/- 28.96%, n = 8), with six of the shifts reaching statistical significance for alpha = 0.05. Because the pattern of torque and EMG responses before and after fatigue suggested the presence of an active force regulator, we used a simple model of the neuromuscular system to estimate a loop gain value for each session. When pre- and postfatigue responses were matched by isometric background torque level, an average loop gain value of 7.9 was computed, whereas for responses matched by average prestretch EMG level, the loop gain estimates averaged 2.1. Although our assessment of force regulation was essentially static and derived from the responses to a single type of perturbation, the change in the incremental torque and EMG stretch responses indicates that meaningful neural compensation for fatigue occurred. Moreover, the loop gain estimates derived from these responses are an order of magnitude larger than those previously reported in animal models, suggesting that force regulation may be important in the control of human muscle contraction.