In a recognition memory experiment we recorded event-related brain potentials (ERPs) following the presentation of faces. We compared correctly classified repetitions of faces (hits) with new faces classified as repeated (false alarms). Stimulus-related averaging yielded significantly more negative ERPs to false alarms between 750 and 900 ms after onset of stimulus presentation. With response-related averaging we found significantly more negative ERPs to false alarms within the last 100 ms before movement onset. The differential activity was distributed over central and frontal regions and was still present in the period after the response. We interpret the phenomenon as an electrophysiological manifestation of false recognition as described by Schacter et al. (Schacter, D.L., Norman, K.A. and Koutstaal, W., The cognitive neuroscience of constructive memory, Annu. Rev. Psychol., 49 (1998) 289-318). False recognition could be the result of a preliminary decision based on illusionary familiarity and be associated with post-retrieval processing.