Objective: To compare the outcome of robot-assisted therapy with dose-matched active control therapy by using accelerometers to study functional recovery in chronic stroke patients. Design: Prospective, randomized, controlled trial. Setting: Stroke units in three medical centres. Subjects: Twenty patients post stroke for a mean of 22 months. Intervention: Robot-assisted therapy ( n = 10) or dose-matched active control therapy ( n = 10). All patients received either of these two therapies for 90–105 minutes each day, 5 days per week, for four weeks. Main measures: Outcome measures included arm activity ratio (the ratio of mean activity between the impaired and unimpaired arm) and scores on the Fugl-Meyer Assessment Scale, Functional Independence Measure, Motor Activity Log and ABILHAND questionnaire. Results: The robot-assisted therapy group significantly increased motor function, hemiplegic arm activity and bilateral arm coordination (Fugl-Meyer Assessment Scale: 51.20 ± 8.82, P = 0.002; mean arm activity ratio: 0.76 ± 0.10, P = 0.026; ABILHAND questionnaire: 1.24 ± 0.28, P = 0.043) compared with the dose-matched active control group (Fugl-Meyer Assessment Scale: 40.90 ± 13.14; mean arm movement ratio: 0.69 ± 0.11; ABILHAND questionnaire: 0.95 ± 0.43). Conclusions: Symmetrical and bilateral robotic practice, combined with functional task training, can significantly improve motor function, arm activity, and self-perceived bilateral arm ability in patients late after stroke.