ERPs were recorded during 5 blocks of a continuous recognition memory task. In each block words were shown twice, separated by an average of 6 intervening items, and the task was to respond discriminatively on the basis of whether each word was appearing in that block for the first (first presentations) or the second (second presentations) time. In blocks 2-5, half of the words had also been shown in the immediately preceding block, and half were new to the experiment. Subjects were slower and less accurate at identifying first presentations of words that had appeared in the preceding block, as compared to first presentations of entirely new words. In the latency ranges of the N400 and late positive components, ERPs to words that had previously been presented were more positive-going than were ERPs to new words. These 'across-block' ERP repetition effects were qualitatively very similar to, and additive with, those evoked by words repeated within a block. These findings suggest that the effects of word repetition on ERPs are not dependent on an invariant mapping between repetition and response and are not a consequence of facilitated decision/response processing for repeated items.