Objective: To compare gait kinetics of persons with chronic hemiparesis who improved their walking after training with a lower-extremity robot virtual reality system with those who trained with a robot alone. Design: Randomized pilot study with a pre-post repeated-measure design. Setting: Motion analysis laboratory in a rehabilitation hospital. Participants: Volunteer sample of 4 community-dwelling persons poststroke. Intervention: The experimental group trained on the robotic device and virtual reality; using ankle movements, subjects navigated a plane or a boat through a virtual environment that contained series of targets. The control group trained using the robotic device alone, without the virtual reality or the haptic feedback. Training included discrete and combined ankle movements performed 3 times a week for 4 weeks. Main Outcome Measures: Kinetic gait measures collected with the Vicon 512 motion capture system. Results: The virtual reality group increased ankle moment (329%), ankle power at push-off (227%), and knee flexion moment at preswing (300%). The control group declined in ankle moment (−28%), power (−12.2%,), and knee flexion moment (55.5%). Conclusions: Improved gait speed in the robot virtual reality group is partially explained by greater ankle moments and power relative to the robot-only group.