Background and Purpose - To compare a computerized arm trainer (AT), allowing repetitive practice of passive and active bilateral forearm and wrist movement cycle, and electromyography-initiated electrical stimulation (ES) of the paretic wrist extensor in severely affected subacute stroke patients. Method - A total of 44 patients, 4 to 8 weeks after stroke causing severe arm paresis (Fugl-Meyer Motor Score [FM, O to 66] <18), were randomly assigned to either AT or ES. All patients practiced 20 minutes every workday for 6 weeks. AT patients performed 800 repetitions per session with the robot and ES patients performed 60 to 80 wrist extensions per session. The primary outcome measure was the blindly assessed FM (0 to 66), and the secondary measures were the upper limb muscle power (Medical Research Council [MRC] sum, 0 to 45) and muscle tone (Ashworth score sum, 0 to 25), assessed at the beginning and end of treatment and at 3-month follow-up. Results - The AT group had a higher Barthel Index score at baseline, but the groups were otherwise homogenous. As expected, FM and MRC sum scores improved overtime in both groups but significantly more in the robot AT group. The initial Barthel Index score had no influence. In the robot AT group, FM score was 15 points higher at study end and 13 points higher at 3-month follow-up than the control ES group. MRC sum score was 15 points higher at study end and at 3-month follow-up compared with the control ES group. Muscle tone remained unchanged, and no side effects occurred. Conclusion - The computerized active arm training produced a superior improvement in upper limb motor control and power compared with ES in severely affected stroke patients. This is probably attributable to the greater number of repetitions and the bilateral approach.