The present study compared neuromuscular fatigue of the knee extensor muscles following highly variable versus constant power output cycling. Ten subjects performed two 33-min cycling trials of the same average power output, in a random order. Cycling exercise was performed either at constant (CST) power output, corresponding to 70% of the maximal aerobic power (MAP), or at variable (VAR) power output with alternating high (200, 150 and 100% of MAP during 10, 15 and 20 s, respectively) and moderate (50% of MAP) power output periods. Neuromuscular tests were performed before and immediately after the two trials. Heart rate (HR) was measured during exercise and blood lactate concentration ([La]) at the end of both trials. Reductions in maximal voluntary contraction torque, voluntary activation level and peak doublet were significantly greater after VAR than after CST. HR and [La] were significantly higher during VAR than during CST. Cycling at a varying power output in comparison to constant power resulted in additional muscular fatigue that may be explained by greater anaerobic contribution and muscle solicitation during the highly variable power output protocol.