Background. Six months after a stroke, the hemiplegic arm often remains compromised. More innovative approaches to motor rehabilitation are needed. Objective. The authors compared a motor learning—based approach in a virtual environment with more conventional upper extremity therapy in a pilot trial. Methods . This prospective, single-blinded, randomized controlled trial compared reinforced feedback in a virtual environment (RFVE; n = 27) with a control intervention (n = 20) of progressive therapy for the affected upper extremity. Both treatments were provided for 4 weeks, 5 days per week, with 1-hour treatment sessions daily. The primary outcomes were the Fugl-Meyer Upper Extremity (F-M UE) and Functional Independence Measure (FIM) scores. Kinematic outcomes included mean duration (MD), mean linear velocity (MLV), and number of submovements to measure the motor performance. Analyses of the primary outcomes were performed per protocol and by intention to treat. Results. F-M UE scores improved significantly in the RFVE group compared with the conventional therapy group (“intention to treat” = 5.10 points, P = .004; ANCOVA = 4.26 points, P < .01). Several of the kinematic parameters improved in the RFVE group (MD, P < .01; MLV, P < .01). FIM improvements did not differ. Conclusions . Both rehabilitation therapies improved arm motor performance and functional activity, but the RFVE therapy induced more robust results in patients exposed to late rehabilitation treatment.