Discontinuous activation of muscle compartments is used to postpone the occurrence of fatigue during both normal activation and artificial stimulation. Periods of force development are interrupted by passive periods during which the muscle can recover. Since it is not known how fatigue parameters depend on intermittent stimulation, we compared fatigue generated by continuous electrical stimulation with fatigue resulting from intermittent stimulation schemes. T5-T6 paraplegics participated in the experiments. Continuous stimulation generated a torque time course that can be described by a rising and a falling exponential time constant. The falling time constant ranged from 100 to 200 sec in all four patients and did not depend on the amplitude of the stimulation pulses. The torque developed at times greater than 250 sec was asymptotically constant and proportional to the maximal torque developed by each patient. Intermittent stimulation appears to postpone fatigue markedly. The torque time course developed in each on/off cycle again was a double exponential. In intermittent stimulation schemes average torque determines the muscle's performance. Therefore, average torque versus time was calculated from the intermittent stimulation data. These relations also follow a double exponential, accurately providing confident estimates of fatigue parameters. Fatigue caused average muscle torque to decline to a constant level which, for each patient, was uncorrelated with the duty cycle of the stimulation pattern. Between patients these levels varied from 18.3 +/- 7.1 to 9.9 +/- 4.3 (% of maximal torque +/- standard error). These findings result in a model that could be of use in controllers for functional neuromuscular stimulation (FNS) that take into account non-stationarity caused by fatigue.