If the quality of care provided by a hospital affects its death rate, then some deaths must be preventable. We have developed a new method to investigate this issue and have reviewed 182 deaths from 12 hospitals (6 high outliers and 6 low outliers for death rate) for three conditions (cerebrovascular accident, pneumonia, or myocardial infarction). The investigators prepared a dictated summary of each patient's hospital course. Then, at least three physicians reviewed each summary and independently judged whether the death could have been prevented. Using a majority rules criterion (at least two of three physicians agreed), we found that 27% of the deaths might have been prevented. Using a unanimity criterion (all three physicians independently agreed), we found a 14% rate of probably preventable deaths. Patients whose deaths were probably preventable were younger (74.7 compared with 78.6 years, P less than 0.05), less often demented (12% compared with 26%, P less than 0.05), and less severely ill (mean Acute Physiology and Chronic Health Evaluation score, 15.6 compared with 21.2; P less than 0.001) than patients whose deaths were nonpreventable. The physicians also listed causes for each probably preventable death; nine reasons encompassed almost all of them. For myocardial infarction, preventable deaths reflected errors in management. For cerebrovascular accident, however, deaths primarily reflected errors in diagnosis. The severity of illness can help a hospital retrospectively identify probably preventable deaths. In the group of patients who died, 42% of those with a low severity of illness had probably preventable deaths as compared to 11% admitted with a high severity of illness. We found that a significant number of hospital deaths might have been prevented. Our findings were based on a new method that needs further testing to substantiate its validity. These findings also need replication before they can be generalized to other hospitals.