A new technique has been developed for identifying, in humans, dynamic spatiotemporal electrical patterns of the brain during purposive behaviors. In this method, single-trial time-series correlations between brain macropotentials recorded from different scalp sites are analyzed by distribution-independent mathematical pattern recognition. Dynamic patterns of correlation clearly distinguished two brief visuomotor tasks differing only in type of mental judgement required (spatial or numeric). These complex patterns shifted in the anterior-posterior and left-right axes between successive 175-millisecond intervals, indicating that many areas in both cerebral hemispheres were involved even in these simple judgements. These patterns were not obtainable by conventional analysis of averaged evoked potentials or by linear analysis of correlations, suggesting that the new technique will advance the study of human brain activity related to cognition and goal-directed behaviors.