BACKGROUND: Autopsy rates have declined worldwide, but recent retrospective intensive care unit (ICU) data indicate major discrepancies between more than 25% of clinical and autopsy diagnoses. METHODS: We conducted a 3-year prospective study of all consecutive autopsies performed on patients who died in a university hospital medical-surgical ICU to determine how many might have benefited from a different level of care, had the autopsy diagnosis been made before death. All clinical diagnoses were compared with autopsy findings at monthly clinical-pathological meetings. Major and minor diagnostic discrepancies were categorized according to the criteria of Goldman et al. RESULTS: Of 1492 patients admitted to the ICU, 315 died, of whom 167 (53.0%) were autopsied. The most common reason (79.7%) for not obtaining an autopsy was family refusal. The mean +/- SD clinical characteristics were similar for autopsied vs nonautopsied patients, except for shorter length of ICU stay (13 +/- 17 vs 20 +/- 27 days, P =.006), shorter duration of mechanical ventilation (13 +/- 16 vs 19 +/- 25 days, P =.01), and lower percentage of postcardiac surgery patients (38.9% vs 50.0%, P =.05). Among the intensivists' 694 clinical diagnoses, 33 (4.8%) were refuted and 13 (1.9%) were judged incomplete by autopsy findings. Autopsies revealed 171 missed diagnoses, including 21 cancers, 12 strokes, 11 myocardial infarctions, 10 pulmonary emboli, and 9 endocarditis, among others. Major diagnostic errors (class I and class II discrepancies) were made in 53 (31.7%) of 167 patients, with a high percentage of immunocompromised patients also observed among these. Similar percentages of patients with class I and class II errors vs other patients had undergone modern diagnostic techniques during their ICU stay. CONCLUSION: Even in the era of modern diagnostic technology, regular comparisons of clinical and autopsy diagnoses provide pertinent information that might improve future management of ICU patients.