OBJECTIVES: We investigate the effect of immediate antiretroviral therapy (ART) eligibility on HIV incidence among HIV-uninfected household members. DESIGN: Regression discontinuity study arising from a population-based cohort. METHODS: Household members of patients seeking care at the Hlabisa HIV Treatment and Care Programme in rural KwaZulu-Natal South Africa between January 2007 and August 2011 with CD4 cell counts up to 350 cells/mul were eligible for inclusion if they had at least two HIV tests and were HIV-uninfected at the time the index patient linked to care (N = 4115). Regression discontinuity was used to assess the intention-to-treat effect of immediate versus delayed ART eligibility on HIV incidence among household members. Exploiting the CD4 cell count-based threshold rule for ART initiation (CD4 < 200 cells/mul until August 2011), we used Cox proportional hazards models to compare outcomes for household members of patients who presented for care with CD4 cell counts just above versus just below the ART initiation threshold. RESULTS: Characteristics of household members of index patients initiating HIV care were balanced between those with an index patient immediately eligible for ART (N = 2489) versus delayed for ART (N = 1626). There were 337 incident HIV infections among household members, corresponding to an HIV incidence of 2.4 infections per 100 person-years (95% confidence interval 2.5-3.1). Immediate eligibility for treatment reduced HIV incidence in households by 47% in our optimal estimate (hazard ratio = 0.53, 95% confidence interval 0.30-0.96), and by 32-60% in alternate specifications of the model. CONCLUSION: Immediate eligibility of ART led to substantial reductions in household-level HIV incidence.