There is considerable evidence to support the hypothesis of relationship between paradoxical sleep (PS) and learning–memory processing. It has been suggested that PS is important in memory retention at the specific time course called PS windows (PSW). The time of PSWs occurrence and duration of these PSWs following the training sessions and, the neurochemical nature of PSWs has not been well known. In this research the duration of PSWs and the effect of caffeine on memory retention in these periods has been investigated. For this purpose, male NMRI rats were trained in a two–way shuttle avoidance task (100 trials in one session) and their memory retention was tested one week after learning. In experiment 1, the baseline EEG and EMG activity 1-4 hr after training was recorded. In those animals that reached to 70% learning criteria (learner group), the total duration of PS increased significantly (P<0.01) 1-4 hr after training. This change was not observed in the non-learner group. In experiment 2, the learner group was made deprived of PS in the two periods of 1-4 hr and 5-8 hr after learning. The memory retention was significantly impaired in 1-4 hr group (P<0.05) but not in the 5-8 hr group. In experiment 3, caffeine (25 mg/kg) was injected i.p 1-4 hr after training to the learner group with or without PS deprivation. Injection of caffeine increased memory significantly (P<0.05), but in PS deprived rats caffeine had no effect on memory retention. In experiment 4, adenosine (7 and 50 mg/kg), physostigmine (0.1 mg/kg) and scopolamine (5 mg/kg) were administered i.p 1-4 hr after training. None of these drugs had a significant effect on memory retention. According to our data, it seems that 1-4 hr period after training could be considered as a PSW in this condition, in which caffeine can increase memory which is not due to cholinergic and adenosine systems. It is concluded that the memory processing that occurs in PSWs could not be attributed to one neurotransmitter system.