During isometric contractions, the fluctuation of the force output of muscles increases as the muscle fatigues, and the contraction is sustained to exhaustion. We analyzed motor unit firing data from the vastus lateralis muscle to investigate which motor unit control parameters were associated with the increased force fluctuation. Subjects performed a sequence of isometric constant-force contractions sustained at 20% maximal force, each spaced by a 6-s rest period. The contractions were performed until the mean value of the force output could not be maintained at the desired level. Intramuscular EMG signals were detected with a quadrifilar fine-wire sensor. The EMG signals were decomposed to identify all of the firings of several motor units by using an artificial intelligence-based set of algorithms. We were able to follow the behavior of the same motor units as the endurance time progressed. The force output of the muscle was filtered to remove contributions from the tracking task. The coefficient of variation of the force was found to increase with endurance time (P < 0.001, R(2) = 0.51). We calculated the coefficient of variation of the firing rates, the synchronization of pairs of motor unit firings, the cross-correlation value of the firing rates of pairs of motor units, the cross-correlation of the firing rates of motor units and the force, and the number of motor units recruited during the contractions. Of these parameters, only the cross-correlation of the firing rates (P < 0.01, R(2) = 0.10) and the number of recruited motor units (P = 0.042, R(2) = 0.22) increased significantly with endurance time for grouped subjects. A significant increase (P < 0.001, R(2) = 0.16) in the cross-correlation of the firing rates and force was also observed. It is suggested that the increase in the cross-correlation of the firing rates is likely due to a decrease in the sensitivity of the proprioceptive feedback from the spindles.