SETTING: Public ambulatory care centers in three districts of northern metropolitan Lima, Peru. OBJECTIVE: To document drug resistance patterns of isolates of Mycobacterium tuberculosis from patients identified as treatment failures under a model tuberculosis (TB) control program based on directly observed, short-course chemotherapy (DOT-SCC). DESIGN: Case series. RESULTS: In a referred, consecutive sample of 173 patients identified as treatment failures on DOT-SCC, 160 (92.5%) had culture-positive TB. Of those 160, 150 (93.8%) had active, pulmonary multidrug-resistant TB (MDR-TB, resistance to at least isoniazid [INH] and rifampicin [RIF]). Sixty of the 150 (40.0%) had isolates resistant to at least INH, RIF, ethambutol (EMB) and pyrazinamide (PZA), the initial first-line empiric treatment regimen used locally. Forty- four (29.3%) had isolates resistant to at least INH, RIF, EMB, PZA and streptomycin (SM), the first retreatment regimen. This series of patients had isolates resistant to a mean of 4.5 of the ten drugs tested. The local profile of multidrug resistance is very different from that obtained from national data from Peru. CONCLUSION: In this setting, treatment failure on DOT-SCC is strongly predictive of active MDR-TB. Because of existing local drug resistance patterns in northern Lima, 89.3% of MDR-TB patients identified as treatment failures will receive ineffective therapy with two or fewer secondary TB drugs if they are given the five-drug empiric retreatment regimen endorsed by the World Health Organization. Further short-course chemotherapy for these patients would only serve to amplify ominous existing drug resistance patterns.