The claim that event-related potentials (ERPs) index familiarity was assessed by acquiring ERPs during a recognition memory task in which participants were instructed to adopt different decision criteria in separate retrieval phases. In one, the instructions were to respond “old” only when confident that this was the correct response, and to respond “new” otherwise (the conservative condition). In the other, the instructions were to respond new only when confident that this was the correct response (the liberal condition). The rationale for this approach was that the level of familiarity licensing an old response would be higher in the conservative than in the liberal condition, and if ERPs index familiarity, this would be reflected in changes to the putative ERP index. This index comprises relatively more positive-going neural activity for correct judgments to old than to new items, which is evident from 300 to 500 msec poststimulus at mid-frontal scalp locations. In keeping with task instructions, participants made more old responses in the liberal than in the conservative condition. There were reliable mid-frontal ERP old/new effects in both conditions, and the ERPs evoked by correct judgments to words in the conservative condition were relatively more positive-going than those in the liberal condition. This finding is consistent with the view that the mid-frontal ERP old/new effect indexes familiarity, and in combination with other ERP findings, provides strong support for dual-process accounts of recognition memory.