A retrospective cohort was set up to identify prognostic factors associated with in-hospital survival in HIV-infected patients admitted to medical intensive care units (MICUs), from 1991 to 1994. Survival from MICU admission to hospital discharge (or in-hospital fatal issue) was estimated and a prognostic score at MICU admission was developed. One hundred and thirty patients were recruited of whom 20% were AIDS-free prior to admission. In-hospital mortality rate was 65%. Median survival was 20 days. The following variables were predictive of mortality: Simplified Acute Physioloy Score II (SAPS II): (hazard ratio [HR]=1.5 for 10 points higher, P<10(-3)), time between HIV diagnosis and admission >5 years (HR=2.7, P<10(-4)), hypoalbuminaemia (HR=1.2 per 5 g/l lower, P=0.03). The prognostic score developed was: SAPS II+25 (if time between HIV diagnosis and MICU admission >5 years) albuminaemia (g/l). A new prognostic score including SAPS II, prior HIV history and albuminaemia better reflected the in-hospital mortality than SAPS II alone. Our findings may still be useful to better evaluate the immediate prognosis of current HIV-infected patients admitted to MICU, particularly those naive to antiretroviral therapy or in treatment failure.