This study was conducted to determine if the source of admission to the medical intensive care unit (MICU) is associated with hospital death independent of the Acute Physiology and Chronic Health Evaluation (APACHE) II score. We calculated APACHE II scores and predicted risks of hospital death for 235 patients who were admitted to the MICU. The predicted death rate was the same as the actual rate for patients who were admitted directly from the emergency department (25% vs 22%), but was less than the actual rate for patients who were transferred from hospital floors (38% vs 55%), the medical intermediate care unit (32% vs 59%), and other hospitals (21% vs 36%). Logistic regression analysis confirmed an independent association between the MICU admission source and risk of death. Our findings suggest that APACHE II does not measure illness severity accurately in all patients who are admitted to intensive care units. If our results are generalizable, using APACHE II to compare intensive care outcomes among hospitals could lead to wrong conclusions about quality of care. Improving predictions of hospital death rates among patients who are in MICUs may require the inclusion of new types of information in the classification system.