BACKGROUND: The diagnostic accuracy of sputum smear microscopy and routine chest radiology for HIV-associated tuberculosis is poor, and culture-based diagnosis is slow, expensive, and is unavailable in most resource-limited settings. We assessed the diagnostic accuracy of a urine antigen test Determine TB-LAM Ag (Determine TB-LAM; Alere, Waltham, MA, USA) for screening for HIV-associated pulmonary tuberculosis before antiretroviral therapy (ART). METHODS: In this descriptive study, consecutive adults referred to a community-based ART clinic in Gugulethu township, South Africa, were all screened for tuberculosis by obtaining sputum samples for fluorescence microscopy, automated liquid culture (gold-standard test), and Xpert MTB/RIF assays (Cepheid, Sunnyvale, CA, USA) and urine samples for the Clearview TB-ELISA (TB-ELISA; Alere, Waltham, MA, USA) and Determine TB-LAM test. Patients with Mycobacterium tuberculosis cultured from one or more sputum samples were defined as cases of tuberculosis. The diagnostic accuracy of Determine TB-LAM used alone or combined with sputum smear microscopy was compared with that of sputum culture and the Xpert MTB/RIF assay for all patients and subgroups of patients stratified by CD4 cell count. FINDINGS: Patients were recruited between March 12, 2010, and April 20, 2011. Of 602 patients enrolled, 542 were able to provide one or more sputum samples, and 94 had culture-positive tuberculosis (prevalence 17·4%, 95% CI 14·2-20·8). Complete results from all tests were available for 516 patients (median CD4 count, 169·5 cells per μL; IQR 100-233), including 85 culture-positive tuberculosis, 24 of whom (28·2%, 95% CI 19·0-39·0) had sputum smear-positive disease. Determine TB-LAM test strips provided results within 30 min. Agreement was very high between two independent readers of the test strips (κ=0·97) and between the test strips and TB-ELISA (κ=0·84). Determine TB-LAM had highest sensitivity at low CD4 cell counts: 66·7% (95% CI 41·0-86·7) at <50 cells per μL, 51·7% (32·5-70·6) at <100 cells per μL, and 39·0% (26·5-52·6) at <200 cells per μL; specificity was greater than 98% for all strata. When combined with smear microscopy (either test positive), sensitivity was 72·2% (95% CI 46·5-90·3) at CD4 counts less than 50 cells per μL, 65·5% (45·7-82·1) at less than 100 cells per μL, and 52·5% (39·1-65·7) at less than 200 cells per μL, which did not differ statistically from the sensitivities obtained by testing a single sputum sample with the Xpert MTB/RIF assay. INTERPRETATION: Determine TB-LAM is a simple, low-cost, alternative to existing diagnostic assays for tuberculosis screening in HIV-infected patients with very low CD4 cell counts and provides important incremental yield when combined with sputum smear microscopy. FUNDING: Wellcome Trust.