Background. We investigated the effects of different intensities of arm and leg rehabilitation training on the functional recovery of activities of daily living (ADL), walking ability, and dexterity of the paretic arm, in a single-blind randomised controlled trial. Methods. Within 14 days after stroke onset, 101 severely disabled patients with a primary middle-cerebral-artery stroke were randomly assigned to: a rehabilitation programme with emphasis on arm training; a rehabilitation programme with emphasis on leg training; or a control programme in which the arm and leg were immobilised with an inflatable pressure splint. Each treatment regimen was applied for 30 min, 5 days a week during the first 20 weeks after stroke. In addition, all patients underwent a basic rehabilitation programme. The main outcome measures were ability in ADL (Barthel index), walking ability (functional ambulation categories), and dexterity of the paretic arm (Action Research aim test) at 6, 12, 20, and 26 weeks. Analyses were by intention to treat. Findings. At week 20, the leg-training group (n = 31) had higher scores than the control group (n = 37) for ADL ability (median 19 [IQR 16-20] vs 16 [10-19], p < 0.05)], walking ability (4 [3-5] vs 3 [1-4], p < 0.05), and dexterity (2 [0-56] vs 0 [0-2], p < 0.01). The arm-training group (n = 33) differed significantly from the control group only in dexterity (9 [0-39] vs 0 [0-2], p < 0.01). There were no significant differences in these endpoints at 20 weeks between the arm-training and leg training groups. Interpretation. Greater intensity of leg rehabilitation improves functional recovery and health-related functional status, whereas greater intensity of arm rehabilitation results in small improvements in dexterity, providing further evidence that exercise therapy primarily induces treatment effects on the abilities at which training is specifically aimed.