Latency to the first episode of rapid eye movement sleep (REML) has been proposed as a potential biomarker for Alzheimer's disease (AD). In this study, we compared REML values from 28 AD patients and 28 age- and sex-matched controls. We employed multiple definitions of REML and multiple cutoffs to classify patients and controls. Results indicated that the best REML definition and optimal cutoff criterion resulted in only 65% correct classifications. We discuss the longer REML in AD patients relative to controls in terms of both overall sleep disturbance and selective deterioration of the REM-cholinergic system. As REML may be relatively short in other forms of psychopathology (e.g., affective disorders), REML may still hold promise in the differential diagnosis of dementia and pseudodementia.