SETTING: Patients with cavitary pulmonary tuberculosis (TB) on baseline chest radiograph (CXR) who remain culture-positive after 8 weeks of treatment are at high risk of relapse. The role of end-of-treatment (EOT) CXR in predicting relapse is unclear. OBJECTIVE: To determine whether EOT CXR independently predicts TB relapse. DESIGN: We conducted a secondary analysis of a randomized trial of intermittent treatment using rifapentine in the continuation phase of TB treatment among 1004 human immunodeficiency virus seronegative adults with culture-proven pulmonary TB. RESULTS: Relapse occurred in 17.3% of subjects with persistent cavity on EOT CXR, in 7.6% of subjects with a cavity that resolved by EOT, and 2.5% (P=0.002 for trend) of subjects who never had a cavity. In multivariable analysis, patients with persistent cavity on EOT CXR were significantly more likely to relapse than patients with no cavity on baseline or 2-month CXR (hazard ratio [HR] 4.22, 95%CI 2.00-8.91), and were more likely to relapse than subjects whose early cavity had resolved by EOT CXR (HR 1.92, 95%CI 1.09-3.39). CONCLUSION: A persistent cavity after 6 months of TB treatment was independently associated with disease relapse after controlling for other variables. EOT CXR may help predict those likely to relapse.