Background The Xpert MTB/RIF test for tuberculosis is being rolled out in many countries, but evidence is lacking regarding its implementation outside laboratories, ability to inform same-day treatment decisions at the point of care, and clinical eff ect on tuberculosis-related morbidity. We aimed to assess the feasibility, accuracy, and clinical eff ect of point-of-care Xpert MTB/RIF testing at primary-care health-care facilities in southern Africa. Methods In this pragmatic, randomised, parallel-group, multicentre trial, we recruited adults with symptoms suggestive of active tuberculosis from fi ve primary-care health-care facilities in South Africa, Zimbabwe, Zambia, and Tanzania. Eligible patients were randomly assigned using pregenerated tables to nurse-performed Xpert MTB/RIF at the clinic or sputum smear microscopy. Participants with a negative test result were empirically managed according to local WHO-compliant guidelines. Our primary outcome was tuberculosis-related morbidity (measured with the TBscore and Karnofsky performance score [KPS]) in culture-positive patients who had begun anti-tuberculosis treatment, measured at 2 months and 6 months after randomisation, analysed by intention to treat. This trial is registered with Clinicaltrials. gov, number NCT01554384. Findings Between April 12, 2011, and March 30, 2012, we randomly assigned 758 patients to smear microscopy (182 culture positive) and 744 to Xpert MTB/RIF (185 culture positive). Median TBscore in culture-positive patients did not diff er between groups at 2 months (2 [IQR 0–3] in the smear microscopy group vs 2 [0·25–3] in the MTB/RIF group; p=0·85) or 6 months (1 [0–3] vs 1 [0–3]; p=0·35), nor did median KPS at 2 months (80 [70–90] vs 90 [80–90]; p=0·23) or 6 months (100 [90–100] vs 100 [90–100]; p=0·85). Point-of-care MTB/RIF had higher sensitivity than microscopy (154 [83%] of 185 vs 91 [50%] of 182; p=0·0001) but similar specifi city (517 [95%] 544 vs 540 [96%] of 560; p=0·25), and had similar sensitivity to laboratory-based MTB/RIF (292 [83%] of 351; p=0·99) but higher specifi city (952 [92%] of 1037; p=0·0173). 34 (5%) of 744 tests with point-of-care MTB/RIF and 82 (6%) of 1411 with laboratory-based MTB/RIF failed (p=0·22). Compared with the microscopy group, more patients in the MTB/RIF group had a same-day diagnosis (178 [24%] of 744 vs 99 [13%] of 758; p<0·0001) and same-day treatment initiation (168 [23%] of 744 vs 115 [15%] of 758; p=0·0002). Although, by end of the study, more culture-positive patients in the MTB/RIF group were on treatment due to reduced dropout (15 [8%] of 185 in the MTB/RIF group did not receive treatment vs 28 [15%] of 182 in the microscopy group; p=0·0302), the proportions of all patients on treatment in each group by day 56 were similar (320 [43%] of 744 in the MTB/RIF group vs 317 [42%] of 758 in the microscopy group; p=0·6408). Interpretation Xpert MTB/RIF can be accurately administered by a nurse in primary-care clinics, resulting in more patients starting same-day treatment, more culture-positive patients starting therapy, and a shorter time to treatment. However, the benefi ts did not translate into lower tuberculosis-related morbidity, partly because of high levels of empirical-evidence-based treatment in smear-negative patients.