Résumé : OBJECTIVE: Verify and explore unexpected results suggesting an effect of deviance direction (shorter or longer deviants) on the amplitude of MMNs evoked by sound duration contrasts. METHODS: MMNs were recorded using the oddball paradigm on ten adults. Four standard stimulus durations (100, 150, 200 and 250ms) were used and deviants were 50% shorter or longer. Behavioral data (hit rates, d', and reaction times) were collected after the electrophysiological sessions. RESULTS: MMNs were larger for short than for long deviants. There was no effect on MMN latencies. Hit rates and d' data were almost at ceiling level for all conditions even for the longest standard - long deviant combination in which the MMN was abolished. CONCLUSIONS: We argue that the deviance direction effect on MMN amplitudes can be explained by the delay between the moment of deviance detection and the end of the deviance quantification process. SIGNIFICANCE: A major effect of deviance direction on amplitudes was confirmed. This effect, which was confined to electrophysiological data, is to be taken into account when using duration contrasts to probe the processing of temporal information.