Visually evoked potentials elicited by tachistoscopically displayed alphabetic stimuli were recorded at vertex from six human subjects during a series of item recognition tasks. When presented with a test stimulus, the observer had to decide whether it had been memorized previously (i.e., whether it belonged to the positive as opposed to the negative set) and made a choice reaction time response accordingly. P300 amplitude was unequivocally greater following positive letter presentations than it was following negative letter presentations. This result is consistent with a “template-matching” model of P30O enhancement, but, more importantly, it extends the potential generality of this theory beyond the detection of simple sensory stimuli to complex cognitive tasks and stimulus events. The latency of the second positive peak remained essentially invariant across all positive set sizes, a fact which suggested that it reflected the minimum time required to register or encode the stimulus input. In contrast, P300 latency increased as positive set size was increased, i.e., as information processing took longer to complete.