BACKGROUND: Universal antiretroviral therapy (ART), as per the 2015 WHO recommendations, might reduce population HIV incidence. We investigated the effect of universal test and treat on HIV acquisition at population level in a high prevalence rural region of South Africa. METHODS: We did a phase 4, open-label, cluster randomised trial of 22 communities in rural KwaZulu-Natal, South Africa. We included individuals residing in the communities who were aged 16 years or older. The clusters were composed of aggregated local areas (neighbourhoods) that had been identified in a previous study in the Hlabisa subdistrict. The study statisticians randomly assigned clusters (1:1) with MapInfo Pro (version 11.0) to either the control or intervention communities, stratified on the basis of antenatal HIV prevalence. We offered residents repeated rapid HIV testing during home-based visits every 6 months for about 4 years in four clusters, 3 years in six clusters, and 2 years in 12 clusters (58 cluster-years) and referred HIV-positive participants to trial clinics for ART (fixed-dose combination of tenofovir, emtricitabine, and efavirenz) regardless of CD4 cell count (intervention) or according to national guidelines (initially </=350 cells per muL and <500 cells per muL from January, 2015; control). Participants and investigators were not masked to treatment allocation. We used dried blood spots once every 6 months provided by participants who were HIV negative at baseline to estimate the primary outcome of HIV incidence with cluster-adjusted Poisson generalised estimated equations in the intention-to-treat population after 58 cluster-years of follow-up. This study is registered with ClinicalTrials.gov, number NCT01509508, and the South African National Clinical Trials Register, number DOH-27-0512-3974. FINDINGS: Between March 9, 2012, and June 30, 2016, we contacted 26 518 (93%) of 28 419 eligible individuals. Of 17 808 (67%) individuals with a first negative dried blood spot test, 14 223 (80%) had subsequent dried blood spot tests, of whom 503 seroconverted after follow-up of 22 891 person-years. Estimated HIV incidence was 2.11 per 100 person-years (95% CI 1.84-2.39) in the intervention group and 2.27 per 100 person-years (2.00-2.54) in the control group (adjusted hazard ratio 1.01, 95% CI 0.87-1.17; p=0.89). We documented one case of suicidal attempt in a woman following HIV seroconversion. 128 patients on ART had 189 life-threatening or grade 4 clinical events: 69 (4%) of 1652 in the control group and 59 (4%) of 1367 in the intervention group (p=0.83). INTERPRETATION: The absence of a lowering of HIV incidence in universal test and treat clusters most likely resulted from poor linkage to care. Policy change to HIV universal test and treat without innovation to improve health access is unlikely to reduce HIV incidence. FUNDING: ANRS, GiZ, and 3ie.