SETTING: Ambulatory, public tuberculosis treatment facilities, central Lima, Peru. OBJECTIVE: To identify risk factors for failure on directly observed Category I therapy. DESIGN: Case-control study. All failures of Category I (2HREZ/4H2R2) therapy in 2000 (2.9% of smear-positive TB patients) were included as cases; two controls per case were matched on health center and approximate time of treatment initiation. RESULTS: The study included 38 cases and 76 controls, all new smear-positive, pulmonary TB patients treated with Category I therapy in central Lima in 2000. Neither treatment irregularity nor incidence of adverse events predicted failure in the study group. Mean baseline body mass index was lower in cases than in controls (P = 0.06). Cases gained less weight during therapy (P = 0.01). Smear positivity at 2 months of therapy was strongly associated with failure (OR 11. 7; 95%CI 2.4-57.5). No controls had positive smears at or after 3 months of therapy (OR [corrected] 144.9; 95%CI 8.4-2500). Nearly 75% of cases with isolates tested for susceptibility to first-line drugs had multidrug-resistant TB (MDR-TB). CONCLUSION: A large proportion of failures on Category I therapy can be identified early. As three-quarters of patients with susceptibility results have MDR-TB, early referral for culture and drug susceptibility testing is critical for prompt initiation of appropriate therapy and improved outcomes.