Résumé : The linear prediction method was applied to obtain a power spectra estimate of sleep EEG in 10 healthy young men. In order to analyze frequency changes of the slow-wave activity (EEG power in the delta band, 0.05-2.88 Hz), the delta mean frequency was computed for each 20-second sleep epoch. A mean delta mean frequency was calculated for each sleep stage in each individual. Our observations indicate that the delta mean frequency decreases during NREM periods and increases mainly at the onset of the REM episodes. This finding, in parallel with the detailed analysis of slow-wave power variations, leads to an additional description of sleep characteristics during transitions between the NREM-REM sleep episodes.