The amplitude and latency of the mismatch negativity (MMN) elicited by occasional shorter-duration tones (25 and 50 ms) in a sequence of 75 ms standard tones were studied in 40 healthy subjects (9-84 years). The replicability and age dependence of the MMN-responses were determined. The 25 ms deviant tone evoked a clear response in 39 of the subjects, while the 50 ms deviant tone evoked an observable MMN only in 32 of the subjects. The MMN peak amplitude for the 25 ms deviants was significantly larger than for the 50 ms deviants. There was no significant difference in the peak latencies (measured from stimulus offset). For the 25 ms deviant, the amplitude diminished with increasing age. The MMN curves for the 25 ms deviant, measured on separate days in 14 subjects, looked very replicable. As a result of noise and filtering effect, the product-moment correlations were poor. The results indicate that the signal-to-noise ratio for the MMN to 25 ms deviants, obtained even in a 25 min recording session, is large enough for clinical use and individual diagnostics when undetectable (or very low amplitude) MMN is used as a sign of pathology. However, judged from the low correlation coefficients, despite the good replicability in visual evaluation, better methods for MMN quantification have to be used for clinical follow-up.