Inpatient length of stay (LOS) is an important measure of hospital activity and is often considered as a proxy of hospital resource consumption. In Portugal, hospitalizations related to HIV infection are some of the most expensive and the second major diagnosis category with greatest average LOS. This paper investigates factors associated with HIV/AIDS LOS. A hierarchical finite normal mixture model was fitted to the logarithm of LOS, to account for the inherent correlation of patients clustered in hospitals. We found that the model with two components had the best fit. In addition associated risk factors were identified for each component and the random effects make possible a comparison of relative efficiencies among hospitals.