The aim of this prospective study was to compare a model based on clinical variables with the clinical judgment of infectious disease specialists to identify HIV-infected patients requiring isolation at admission in order to prevent the nosocomial transmission of tuberculosis. Clinical, epidemiological and radiological variables available at admission were recorded for 362 admissions of 274 HIV-infected patients. Using multiple logistic regression analysis, a model to identify patients with tuberculosis was developed based on four clinical variables (node enlargement, constitutional symptoms, intravenous drug use, history of previous correct therapy for tuberculosis) and a positive auramine sputum stain. This model was applied to each of the 362 admissions studied. The decision made by the infectious disease specialist at admission was also recorded. The results indicate that application of the model would have allowed physicians to correctly identify and isolate 24 of 27 patients with tuberculosis, while 5.4 patients without tuberculosis would have been unnecessarily isolated for every patient with tuberculosis. The results for the infectious disease specialists were slightly better, with 26 of 27 patients with tuberculosis being identified and isolated correctly and only 3.2 patients being isolated unnecessarily for every patient with tuberculosis. Thus, a simple model based on clinical variables may be useful in helping physicians identify tuberculosis carriers among HIV-infected patients, but infectious disease specialists are able to identify them more efficiently.