Most patients who have had a stroke are dependent on informal caregivers for activities of daily living. The TRACS trial investigated a training programme for caregivers (the London Stroke Carers Training Course, LSCTC) on physical and psychological outcomes, including cost-effectiveness, for patients and caregivers after a disabling stroke. We undertook a pragmatic, multicentre, cluster randomised controlled trial with a parallel cost-effectiveness analysis. Stroke units were eligible if four of five criteria used to define a stroke unit were met, a substantial number of patients on the unit had a diagnosis of stroke, staff were able to deliver the LSCTC, and most patients were discharged to a permanent place of residence. Stroke units were randomly assigned to either LSCTC or usual care (control group), stratified by geographical region and quality of care, and using blocks of size 2. Patients with a diagnosis of stroke, likely to return home with residual disability and with a caregiver providing support were eligible. The primary outcome for patients was self-reported extended activities of daily living at 6 months, measured with the Nottingham Extended Activities of Daily Living (NEADL) scale. The primary outcome for caregivers was self-reported burden at 6 months, measured with the caregivers burden scale (CBS). We combined patient and caregiver costs with primary outcomes and quality-adjusted life-years (QALYs) to assess cost-effectiveness. This trial is registered with controlled-trials.com, number ISRCTN 49208824. We assessed 49 stroke units for eligibility, of which 36 were randomly assigned to either the intervention group or the control group. Between Feb 27, 2008, and Feb 9, 2010, 928 patient and caregiver dyads were registered, of which 450 were in the intervention group, and 478 in the control group. Patients' self-reported extended activities of daily living did not differ between groups at 6 months (adjusted mean NEADL score 27·4 in the intervention group versus 27·6 in the control group, difference -0·2 points [95% CI -3·0 to 2·5], p value=0·866, ICC=0·027). The caregiver burden scale did not differ between groups either (adjusted mean CBS 45·5 in the intervention group versus 45·0 in the control group, difference 0·5 points [95% CI -1·7 to 2·7], p value=0·660, ICC=0·013). Patient and caregiver costs were similar in both groups (length of the initial stroke admission and associated costs were £13,127 for the intervention group and £12,471 for the control group; adjusted mean difference £1243 [95% CI -1533 to 4019]; p value=0·380). Probabilities of cost-effectiveness based on QALYs were low. In a large scale, robust evaluation, results from this study have shown no differences between the LSCTC and usual care on any of the assessed outcomes. The immediate period after stroke might not be the ideal time to deliver structured caregiver training. Medical Research Council.