Using electroencephalographic methods, rats learning or not learning a two-way active avoidance task were found to differ significantly in the structure of sleep determined the day before training. The main differences concerned (i) synchronized sleep episodes followed by wakefulness, which were longer and fewer in learning rats; (ii) paradoxical sleep episodes, which were longer in learning rats. Significant correlations were present between the number and/or the average duration of synchronized sleep episodes followed by wakefulness or by paradoxical sleep and the number of avoidances or escapes scored in the training session. Power spectral analysis indicated that the relative output in the 6 – 7-Hz region was higher in learning rats, notably during short episodes of synchronized sleep followed by paradoxical sleep. As two-way active avoidance training induces comparable modifications in postacquisition sleep (Ambrosini et al., Physiol. Behav., 51, 217 – 226, 1992), the features of preacquisition sleep which prevail in learning rats might directly determine their capacity to learn. Alternatively, they might reflect the existence of a genetic determinant independently conditioning the ability to learn.