We randomly assigned frail elderly inpatients with a high probability of nursing-home placement to an innovative geriatric evaluation unit intended to provide improved diagnostic assessment, therapy, rehabilitation, and placement. Patients randomly assigned to the experimental (n = 63) and control (n = 60) groups were equivalent at entry. At one year, patients who had been assigned to the geriatric unit had much lower mortality than controls (23.8 vs. 48.3 per cent, P less than 0.005) and were less likely to have initially been discharged to a nursing home (12.7 vs. 30.0 per cent, P less than 0.05) or to have spent any time in nursing home during the follow-up period (26.9 vs. 46.7 per cent, P less than 0.05). The control-group patients had substantially more acute-care hospital days, nursing-home days, and acute-care hospital readmissions. Patients in the geriatric unit were significantly more likely to have improvement in functional status and morale than controls (P less than 0.05). Direct costs for institutional care were lower for the experimental group, especially after adjustment for survival. We conclude that geriatric evaluation units can provide substantial benefits at minimal cost for appropriate groups of elderly patients, over and above the benefits of traditional hospital approaches.