In a prospective study of EEG sleep patterns in 25 elderly depressives, 25 elderly demented patients, and 25 healthy, elderly control subjects, the sleep of depressives was characterized by reduced REM sleep latency, increased REM percent and first REM period density, and altered temporal distribution of REM sleep, as well as by diminished sleep maintenance (correlated significantly with Hamilton ratings of depression: multiple R = -0.42, p less than 0.05). In contrast, the sleep of demented patients showed reduced REM sleep percent, but normal REM temporal distribution, increased loss of spindles and K-complexes (the latter correlating significantly with severity of cognitive impairment as measured by the Folstein score: multiple R = -0.59, p less than 0.01), and less severe sleep maintenance difficulty than for depressives. An examination of REM latency demonstrated a skewed distribution in depression (i.e., 42% of nights with sleep-onset REM periods), but a normal distribution in the controls and demented subjects. A REM latency cut-off score of 30 min correctly classified 68% of all patients (kappa = 0.36; p less than 0.005), compared with 78% correctly identified in our retrospective study (Reynolds et al. 1983).