Neural synchronization in the cortex, and its potential role in information coding, has attracted much recent attention. In this study, we have recorded long spike trains (mean, 33,000 spikes) simultaneously from multiple single neurons in the primary motor cortex (M1) of two conscious macaque monkeys performing a precision grip task. The task required the monkey to use its index finger and thumb to move two spring-loaded levers into a target, hold them there for 1 s, and release for a food reward. Synchrony was analyzed using a time-resolved cross-correlation method, normalized using an estimate of the instantaneous firing rate of the cell. This was shown to be more reliable than methods using trial-averaged firing rate. A total of 375 neurons was recorded from the M1 hand area; 235 were identified as pyramidal tract neurons. Synchrony was weak [mean k' = 1.05 +/- 0.04 (SD)] but widespread among pairs of M1 neurons (218/1359 pairs with above-chance synchrony), including output neurons. Synchrony usually took the form of a broad central peak [average width, 18.7 +/- 8.7 (SD) ms]. There were marked changes during different phases of the task. As a population, synchrony was greatest during the steady hold period in striking contrast to the averaged cell firing rate, which was maximal when the animal was moving the levers into target. However, the modulation of synchrony during task performance showed considerable variation across individual cell pairs. Two types of synchrony were identified: oscillatory (with periodic side lobes in the cross-correlation) and nonoscillatory. Their relative contributions were quantified by filtering the cross-correlations to exclude either frequencies from 18 to 37 Hz or all higher and lower frequencies. At the peak of population synchrony during the hold period, about half (51.7% in one monkey, 56.2% in the other) of the synchronization was within this oscillatory bandwidth. This study provides strong support for assemblies of neurons being synchronized during specific phases of a complex task with potentially important consequences for both information processing within M1 and for the impact of M1 commands on target motoneurons.