BACKGROUND: Children and adolescents have poorer HIV treatment outcomes than adults. We aimed to assess the effect of community-based support for caregivers of HIV-infected children and adolescents, who are key mediators to children engaging with care, on treatment outcomes. METHODS: In this open-label, randomised contolled trial, we recruited children aged 6-15 years with newly-diagnosed HIV attending primary health-care clinics in Harare, Zimbabwe. Children were randomly assigned to receive decentralised primary health-care clinic-based HIV care (control group), according to national guidelines for 18 months, or decentralised care plus structured support visits by trained community health workers (intervention group) according to national guidelines for 18 months. Primary outcomes were the proportion of participants who died or had an HIV viral load of 400 copies per mL or higher at 12 months after antiretroviral therapy (ART) initiation (among those who started ART within 6 months of enrolment); and the proportion who missed two or more scheduled clinic visits by 18 months post-enrolment (among all participants). Analyses were complete-case, modified-intention-to-treat. This trial is registered with the Pan African Clinical Trials Registry, number PACTR201212000442288. FINDINGS: Between January, 2013, and January, 2015, 470 participants tested HIV-positive at seven study primary health-care clinics and were screened for eligibility. Of the 334 eligible children and adolescents, 166 were randomly assigned to the intervention group and 168 to the control group. The median age of participants was 11 years (IQR 8-13) and 178 (53%) were girls. Among the 238 participants who started ART within 6 months of enrolment, the proportion who died or had a viral load of 400 copies/mL or higher at 12 months post-ART initiation was significantly lower in the intervention group than in the control group (31 [33%] of 94 participants vs 42 [49%] of 86 participants, respectively, adjusted odds ratio [aOR] 0·46, 95% CI 0·23-0·89; p=0·02). The proportion of children missing two or more scheduled visits was similar in the intervention group and control group (27 [17%] of 155 vs 30 [18%] of 165, aOR 0·92, 95% CI 0·49-1·74; p=0·79). One participant withdrew from the trial 240 days after enrolment and 12 died during follow-up (five in the intervention group; seven in the control group). INTERPRETATION: Community-based support for caregivers has high potential for scalability and could have a substantial effect on HIV virological suppression in children and adolescents, a group with disproportionately poor treatment outcomes. FUNDING: Wellcome Trust.