Event-related potentials (ERPs) were recorded while subjects performed 2 recognition memory tasks. In the first task, a series of words was presented in which items were repeated after either 6 or 19 intervening words. Subjects were required to make an 'old/new' discrimination to every word. In the second task, which was performed approximately 45 min after the first, subjects had to discriminate between words employed in the first task and previously unpresented items. In the first task, ERPs to 'old' words were more positive-going, from around 250 msec post stimulus, than those to 'new' items. Old/new ERP differences in the second task were smaller than in the first, less widespread over the scalp, and had an onset latency of around 500 msec, although subjects' ability to make 'old/new' discriminations was at a high level. It is suggested that 'old/new' ERP differences may not reflect processes necessary for discrimination between old and new items in recognition memory tasks.