SETTING: Mortality associated with human immunodeficiency virus (HIV) related multidrug-resistant tuberculosis (MDR-TB) is reduced with effective early therapy. Identifying predictors of, and effective regimens for, MDR-TB is critical. OBJECTIVE: A multicenter prospective study was initiated to 1) determine the demographic, behavioral, clinical and geographic risk factors associated with the occurrence of MDR-TB among HIV-infected patients, and 2) to evaluate the overall survival and clinical response of MDR-TB patients treated with specific drug regimens. METHODS: Patients were prospectively evaluated for MDR-TB. Information included history of prior treatment for tuberculosis, close contact with a known case of MDR-TB, and residence in a facility with known or suspected MDR-TB transmission. Patients with known MDR-TB, or those suspected to be at high risk, were offered enrollment in a treatment pilot study. Study drugs included levofloxacin and at least two additional drugs to which the patient's isolate was known, or most likely, to be susceptible. Survival was the primary endpoint. RESULTS: Complete data are available for 156 HIV-infected patients with confirmed tuberculosis. Sixteen (10%) had MDR-TB. Only a history of prior tuberculosis treatment was associated with MDR-TB in multivariate analysis (OR = 4.4, P < 0.02). Twelve patients with MDR-TB enrolled in the treatment pilot had a median CD4 cell count of 51/mm3. The cumulative probability of survival at one year was 75% (95% CI 50.5-99.5) and at 18 months, 65.6% (95% CI 38.1-93.1). Toxicity requiring discontinuation of medications occurred in two patients. CONCLUSIONS: A history of treatment for tuberculosis was the only predictor for MDR-TB in a cohort of HIV-infected patients with tuberculosis. In addition, this prospective study supports the results of prior retrospective studies that effective treatment impacts on mortality. Current second-line treatment, including high dose levofloxacin, appears to be reasonably well tolerated.