Voluntary movements are thought to be controlled via a well-ordered, spatially discrete, somatotopic map in the primary motor cortex (M1). We examined this hypothesis in monkeys trained to perform visually-cued, individuated flexion and extension movements of each digit and of the wrist. Single neurone recordings in M1 during such finger movements revealed two unexpected features. First, single M1 neurones often discharge during instructed movements of multiple digits. Second, neurones active during any particular instructed movement are distributed widely throughout the same M1 territory as neurones active during any other movement. Reversible, partial inactivation of the M1 hand representation produced by injection of 5-10 microg muscimol at one site impaired the monkeys' ability to perform finger movements, but no relationship was evident between the particular finger movements that were affected and the mediolateral location of the injection site along the central sulcus. Thus each finger movement is represented by activity distributed widely in the M1 upper extremity representation. If not controlled from spatially segregated M1 regions, movements of different fingers might be controlled by groups of spatially scattered but physiologically similar neurones. Cluster analysis of M1 neurones demonstrated a large group that discharged during most finger movements, and a small group that paused during most movements. Distinct functional groups of M1 neurones that might control particular finger movements were identified inconsistently. We therefore hypothesize that M1 neurones are a very diverse network controlling finger movements.