The event-related potential technique was used to investigate the time course and scalp-potential topography for the numerical distance effect in a number-matching task. Twenty undergraduates judged whether a number matched or did not match another number presented 1.5 s earlier. Compared with number pairs with a 'small' numerical distance (distance=1), number pairs with a 'large' numerical distance (greater than 2) showed a longer latency and a greater positive amplitude in the N240 component. This numerical distance effect was limited to fronto-central electrodes. These results were discussed in terms of the neural bases of the numerical distance effect during the automatic processing of numbers.