Objective: To investigate the immediate response to avatar-based biofeedback on 3 clinically important gait parameters: step length, knee extension, and ankle power in children with cerebral palsy (CP). Design: Repeated measures design. Setting: Rehabilitation clinic. Participants: Children with spastic paresis (N=22; 10.5±3.1y), able to walk without assistive devices. Intervention: Children walked on a treadmill with a virtual reality environment. Following baseline gait analysis, they were challenged to improve aspects of gait. Children visualized themselves as an avatar, representing movement in real time. They underwent a series of 2-minute trials receiving avatar-based biofeedback on step length, knee extension, and ankle power. To investigate optimization of biofeedback visualization, additional trials in which knee extension was visualized as a simple bar with no avatar; and avatar alone with no specific biofeedback were carried out. Main Outcome Measures: Gait pattern, as measured by joint angles, powers, and spatiotemporal parameters, were compared between baseline and biofeedback trials. Results: Participants were able to adapt gait pattern with biofeedback, in an immediate response, reaching large increases in ankle power generation at push-off (37.7%) and clinically important improvements in knee extension (7.4^o) and step length (12.7%). Biofeedback on one parameter had indirect influence on other aspects of gait. Conclusion: Children with CP show capacity in motor function to achieve improvements in clinically important aspects of gait. Visualizing biofeedback with an avatar was subjectively preferential compared to a simplified bar presentation of knee angle. Future studies are required to investigate if observed transient effects of biofeedback can be retained with prolonged training to test whether biofeedback-based gait training may be implemented as a therapy tool.