1. Voluntary activation of elbow flexor muscles can be optimal during brief maximal voluntary contractions (MVCs), although central fatigue, a progressive decline in the ability to drive the muscle maximally, develops during sustained or repeated efforts. We stimulated the motor cortex and motor point in human subjects to investigate motor output during fatigue. 2. The increment in force (relative to the voluntary force) produced by stimulation of the motor point of biceps brachii increased during sustained isometric MVCs of the elbow flexors. Motoneuronal output became suboptimal during the contraction, i.e. central fatigue developed and accounted for a small but significant loss of maximal voluntary force. During 3 min MVCs, voluntary activation of biceps fell to an average of 90.7% from an average of > 99%. 3. The increment in force (relative to the voluntary force) produced by magnetic cortical stimulation was initially small (1.0%) but also increased during sustained MVCs to 9.8% (with a 2 min MVC). Thus, cortical output was not optimal at the time of stimulation nor were sites distal to the motor cortex already acting maximally. 4. A sphygmomanometer cuff around the upper arm blocked blood supply to brachioradialis near the end of a sustained MVC and throughout subsequent brief MVCs. Neither maximal voluntary force nor voluntary activation recovered during ischaemia after the sustained MVC. However, fatigue-induced changes in EMG responses to magnetic cortical stimulation recovered rapidly despite maintained ischaemia. 5. In conclusion, during sustained MVCs, voluntary activation becomes less than optimal so that force can be increased by stimulation of the motor cortex or the motor nerve. Complex changes in excitability of the motor cortex also occur with fatigue, but can be dissociated from the impairment of voluntary activation. We argue that inadequate neural drive effectively 'upstream' of the motor cortex must be one site involved in the genesis of central fatigue.