Sleep measurements as determined by all-night electroencephalogram recordings and the sleep ratings by the nursing staff were compared in 20 patients during periods of behavioral and psychological disturbance. In all 7 schizophrenic patients actual sleep was accurately determined by the nurses. Only in 4 the 12 depressed patients did actual sleep comparisons achieve high nursing reliability. Intermittent wakefulness assessed by matched pairs suggested that nurses had difficulty distinguishing wakefulness from sleep in depressed patients. Using the duration of total sleep comparisons failed to improve nursing reliability with depressed patients. Both sleep latency and early morning awakening were accurately determined in less than 25% of the cases. These results point up the need to re-evaluate previous sleep studies dependent on observational data, especially those relating to depression.