Background Most people with stroke in India have no access to organised rehabilitation services. The effectiveness of training family members to provide stroke rehabilitation is uncertain. Our primary objective was to determine whether family-led stroke rehabilitation, initiated in hospital and continued at home, would be superior to usual care in a low-resource setting. Methods The Family-led Rehabilitation after Stroke in India (ATTEND) trial was a prospectively randomised open trial with blinded endpoint done across 14 hospitals in India. Patients aged 18 years or older who had had a stroke within the past month, had residual disability and reasonable expectation of survival, and who had an informal family-nominated caregiver were randomly assigned to intervention or usual care by site coordinators using a secure web-based system with minimisation by site and stroke severity. The family members of participants in the intervention group received additional structured rehabilitation training—including information provision, joint goal setting, carer training, and task-specific training—that was started in hospital and continued at home for up to 2 months. The primary outcome was death or dependency at 6 months, defined by scores 3–6 on the modified Rankin scale (range, 0 [no symptoms] to 6 [death]) as assessed by masked observers. Analyses were by intention to treat. This trial is registered with Clinical Trials Registry-India (CTRI/2013/04/003557), Australian New Zealand Clinical Trials Registry (ACTRN12613000078752), and Universal Trial Number (U1111-1138-6707). Findings Between Jan 13, 2014, and Feb 12, 2016, 1250 patients were randomly assigned to intervention (n=623) or control (n=627) groups. 33 patients were lost to follow-up (14 intervention, 19 control) and five patients withdrew (two intervention, three control). At 6 months, 285 (47%) of 607 patients in the intervention group and 287 (47%) of 605 controls were dead or dependent (odds ratio 0·98, 95% CI 0·78–1·23, p=0·87). 72 (12%) patients in the intervention group and 86 (14%) in the control group died (p=0·27), and we observed no difference in rehospitalisation (89 [14%]patients in the intervention group vs 82 [13%] in the control group; p=0·56). We also found no difference in total non-fatal events (112 events in 82 [13%] intervention patients vs 110 events in 79 [13%] control patients; p=0·80). Interpretation Although task shifting is an attractive solution for health-care sustainability, our results do not support investment in new stroke rehabilitation services that shift tasks to family caregivers, unless new evidence emerges. A future avenue of research should be to investigate the effects of task shifting to health-care assistants or team-based community care. Funding The National Health and Medical Research Council of Australia.