This article discusses an analytical approach for interpreting hospital death rates. A study of patient mortality in four hospitals in a large metropolitan area demonstrates the principles underlying this approach. One hospital, which experienced an annual crude death rate almost twice that of the other three, was suspected of providing an inferior quality of care. Case-mix differences among the hospitals introduced a serious bias, however, and were later taken into account. The patients' primary diagnosis was found to be the most important case-mix variable, with a potential for biasing death rate comparisons. After adjustment for case mix, the maximum difference in death rates among the four hospitals was reduced from 19 to four deaths per 1,000 patients. Further analysis of diagnosis-specific mortality supported the thesis that the observed mortality excess was largely attributable to patient referral patterns in the community.