Using electroencephalographic sleep data from a sample of 235 elderly subjects, discriminant function analyses of sleep alterations in depression and dementia were performed. Overall, 80% of patients were correctly classified using a backward discriminant function analysis, and 81% with a general stepwise discriminant function analysis. Four measures contributed to the separation of depressed and demented patients: rapid eye movement (REM) sleep latency (lower in depressives); REM sleep percent (higher in depressives); indeterminate non-REM sleep percent (higher in demented patients, reflecting greater loss of spindles and K complexes); and early morning awakening (more marked in depressives). When both discriminant functions were subjected to cross-validation in independent subsamples, both procedures correctly identified 78% of patients. The classification functions derived from nondemented depressed and nondepressed demented patients were applied to a mixed-symptom group (n = 42). Overall, 27 patients (64%) with either depressive pseudodementia or dementia with depressive features were correctly classified using the same four predictor variables. These findings suggest that sleep physiological alterations of depression and dementia reflect between-group differences in sleep continuity, sleep architecture, and REM sleep temporal distribution, and that the differences are statistically reliable, in both diagnostically pure and mixed clinical presentations. These findings are discussed in the context of current hypotheses of sleep regulation and its mechanisms.