Recent studies indicated a selective activation during rapid eye movement (REM) sleep of the amygdala known to play a decisive role in the processing of emotional stimuli. This study compared memory retention of emotional versus neutral text material over intervals covering either early sleep known to be dominated by nonREM slow wave sleep (SWS) or late sleep, in which REM sleep is dominant. Two groups of men were tested across 3-h periods of early and late sleep (sleep group) or corresponding retention intervals filled with wakefulness (wake group). Sleep was recorded polysomnographically. Cortisol concentrations in saliva were monitored at acquisition and retrieval testing. As expected, the amount of REM sleep was about three times greater during late than during early retention sleep, whereas a reversed pattern was observed for SWS distribution (P < 0.001). Sleep improved retention, compared with the effects of wake intervals (P < 0.02). However, this effect was substantial only in the late night (P < 0.005), during which retention was generally worse than during the early night (P < 0.02). Late sleep particularly enhanced memory for emotional texts. This effect was highly significant in comparison with memory for neutral texts (P < 0.01) and in comparison with memory after late and early wake intervals (P < 0.001). Cortisol concentration differed between early and late retention intervals but not between sleep and wake conditions. Results are consonant with a supportive function of REM sleep predominating late sleep for the formation of emotional memory in humans.It is well established that sleep has a facilitating effect on memory formation in animals and humans (e.g., Jenkins and Dallenbach 1924;Smith 1985Smith , 1995Stickgold et al. 2000). However, the physiological mechanisms of this effect are still unclear. Specifically, there is an ongoing controversy concerning the differential contribution of rapid eye movement (REM) sleep and slow wave sleep (SWS) to memory consolidation. In particular, human studies provided mixed results, some indicating a beneficial effect of REM sleep on memory (Empson and Clarke 1970;Lewin and Glaubman 1975;Tilley and Empson 1978) whereas others suggested SWS to be critical for memory formation (Yaroush et al. 1971;Barrett and Ekstrand 1972;Fowler et al. 1973;Ekstrand 1977). The divergent outcomes appear to reflect differences in the methods used for investigating memory formation during sleep and in the type of memory examined. The most commonly used approach in the study of sleepassociated memory formation is the REM sleep deprivation paradigm, in which subjects are awakened whenever polysomnographic recordings indicate signs of REM sleep. In control conditions, awakenings are induced in nonREM sleep. Yet the REM sleep-deprivation paradigm has been subject to profound criticism (Horne and McGrath 1984; Vertes and Eastman, in press). Compared with other sleep stages, frequent arousal from REM sleep induces serious emotional and cognitive disturbances. These disturbances impa...