We identify early predictors of multidrug-resistant tuberculosis and describe improved clinical outcomes, including survival, for patients with human immunodeficiency virus (HIV)—related multidrug-resistant tuberculosis (MDR-TB) when they are prospectively identified and receive treatment under direct observation. Analysis by means of a Cox proportional hazards model revealed that failure to defervesce while receiving a standard four-drug antituberculous regimen was independently associated with multidrug resistance (P = .004). When patients with HIV-related MDR-TB were prospectively identified and treated with at least two agents that were active in vitro, 100% bacteriologic conversion and improved survival (⩾4 months for 88% of patients and ⩾1 year for 59% of patients) were observed. For patients with HIV-related tuberculosis, poorer survival was associated with a CD4+ lymphocyte count of 3 (P = .03); multidrug resistance was not a predictor of poor outcome (P = .82). These data suggest that patients with prolonged fever who are receiving antituberculous therapy may be an appropriate subgroup to target for broader empirical therapy. The findings also demonstrate that improved outcomes can be achieved with HIV-related MDR-TB when patients are prospectively identified and treated with agents that are active in vitro.