Thirty-two narcoleptic subjects with excessive daytime sleepiness and cataplexy were recorded for 33 continuous hours. The continuous polysomnographic recording (CPSG) was followed by a standard MSLT at 2-h intervals. There were 64 sleep onset REM episodes (SOREMs) vs 64 sleep onset nonREM episodes (SONREMs) during the CPSG, and 102 SOREMs vs 50 SONREMS during the MSLT. Both sleep onset types peaked at 13-15 h during the CPSG while sleep onsets were evenly distributed during the MSLT. In the latter procedure, the mean sleep latency was significantly shorter with SOREMs occurrence than with SONREMs occurrence. Two factors were extracted in each procedure by means of a Varimax Rotated Factor Analysis. During the CPSG, SOREMs were related to the preceding nocturnal sleep parameters in the first factor, and to the daytime total sleep time and the total number of sleep onsets in the second factor. During the MSLT, SOREMs were related only to the mean sleep latency and the total number of sleep onsets. It was concluded that the occurrence of SOREMs is primarily due to the residual somnolence in narcoleptic subjects. However, their occurrence during the MSLT is largely independent of the prior history of sleep and waking. Thus, we propose a nonREM-REM sleep dysregulation hypothesis to account for the appearance of SOREMs in narcolepsy.