Long-latency auditory evoked potentials were recorded from two groups of patients, with and without dementia, and were compared with those from a population of normal subjects ranging in age from 15 to 76 years. A sequence of tones of two different frequencies (1000 Hz and 2000 Hz) was presented and each patient was asked to count the occurrences of the rare (P = 0.15) tones in the sequence. Evoked potential waveforms were averaged separately for the rare and frequent tone. Of the various evoked potential components elicited by the tones, the P3 component (latency 300-500 ms) was found to be the most sensitive to aging in normal subjects. It was also the only component which could be used to differentiate between the demented patients and the normal subjects or non-demented patients. The non-demented patients did not differ from normal in any waveform measure. The magnitude of the latency change of the P3 component in dementia relative to normal was sufficiently large that it may provide a practical and objective measure of dementia in a clinical setting.