This study tested the two main neurocognitive models of dreaming by using cognitive data elicited from REM sleep in normals and narcoleptics. The two models were the ''activation-only'' view which holds that, in the context of sleep, overall activation of the brain is sufficient for consciousness to proceed in the manner of dreaming (e.g., Antrobus, 1991;Foulkes, 1993;Vogel, 1978); and the Activation, Input source, Modulation (AIM model), which predicts that not only brain activation level but also neurochemical modulatory systems exert widespread effects upon dreaming (Hobson & McCarley, 1977;Hobson, Pace-Schott, & Stickgold, 2000). Mental activity was studied in nocturnal REM in 15 narcoleptics and 9 normal healthy persons and in REM at the onset of daytime naps and nighttime sleep (SOREM) in narcoleptics. The study was performed in the subjects' homes, using instrumental awakenings and ambulatory polysomnographic techniques, and focused upon visual vividness, mentation report length, improbable and discontinuous bizarre features, and reflective consciousness. Within each subject group, most cognitive variables tended to fluctuate in line with expected variations in circadian activation level. When comparing the cognitive variables between the two groups, reflective consciousness was clearly highest in narcoleptics, whereas improbabilities and discontinuities were lower, with mentation report length and visual vividness differing less between the groups. These findings are consistent with the AIM model of sleep mentation, but not with the activation-only model.