To contrast mortality for groups of ICU patients treated in different hospitals, we surveyed 795 consecutive ICU admissions in 5 ICUs using a general severity of illness classification system. After obtaining information from the medical record on age, sex, indication for ICU admission, and severity of illness, we used a logistic multiple regression equation to project death rates for each ICU based on data from a sixth reference hospital. There were substantial differences in severity of acute illness among the hospitals which accounted for most of the variation in death rates. In all ICUs, however, projected death rates were quite similar to observed deaths. These findings suggest that the use of a general severity of illness index and multivariate statistical techniques could, after further refinement and validation, improve interhospital comparisons of the outcome of acutely ill patients.