Event-related potentials (ERPs) were recorded during a continuous recognition memory task for visually presented words, with a 6 item lag between the first and second presentation of each word. The subjects consisted of: (i) a control group of patients with primary generalized epilepsy; (ii) patients who had undergone either a left or a right anterior temporal lobectomy; (iii) unoperated patients with either left or right unilateral temporal lobe epilepsy. In the controls, ERPs to detected 'old' words were reliably more positive-going in the interval 300 to 600 ms post-stimulus than were ERPs to 'new' items. In both left and right lobectomy patients, 'old/new' ERP differences in the same latency range were significantly smaller than in the controls, and did not differ significantly from zero. At midline electrodes, old/new effects in the temporal lobe epilepsy patients were of similar magnitude to those of the controls. In contrast to the control data, the old/new effects in the temporal lobe epilepsy patients were asymmetric, in that they were smaller over the hemisphere ipsilateral to the seizure focus than over the contralateral hemisphere. No relationship was found across subjects between the magnitude of old/new ERP effects and verbal memory performance. In a second task, occasional non-words had to be discriminated against a background of sequentially presented words, some of which were repetitions of the immediately preceding item. ERPs evoked by repeated words were more positive-going than were those to first presentations; this effect was reliable, and of equivalent size, in all patient groups. It is concluded that in the recognition task, old/new ERP effects are dependent on temporal lobe functioning, but that the anterior temporal lobe is not the principal locus of the generators of these effects. The cognitive processes reflected by these effects do not appear to be strongly lateralized to one hemisphere, and neither do they seem to be necessary for normal verbal memory.