In two recognition memory tests subjects made initial old/new judgements and subsequently judged whether 'old' words had been presented auditorily or visually at study. Test words were presented visually in Experiment 1 and auditorily in Experiment 2. ERPs evoked at test by words correctly judged 'old' were separated according to whether they were correctly or incorrectly assigned to their study modality. In both experiments, ERPs to correctly assigned words were more positive than those evoked by words correctly judged to be new. This ERP old/new effect was absent for words incorrectly assigned to study modality in Experiment 1, and was observed over a restricted latency range in Experiment 2. The findings are consistent with the hypothesis that the old/new effect is associated with recognition memory based on recollection of the study episode, rather than familiarity.