Project Summary Recent large randomized clinical trials targeting rehabilitation of arm and hand function after stroke have failed to demonstrate superiority of specific task-oriented rehabilitation interventions. Here, based on theoretical perspectives in error-based motor learning, two previous studies from our laboratories, and our clinical experience, we argue that a possible cause of failures of these trials is that practice movements were not performed with sufficient speed during training. Our general hypothesis is that intensive “speed training”, that is, a type of training that emphasizes many repetitions of fast arm movements in a skill learning task, is effective in improving the recovery of arm movements in individuals with mild to moderate motor impairment post-stroke. Importantly, such training promotes re-acquisition of feedforward control that is primarily mediated through cerebellar processes. We have three aims for this high-risk high-reward theoretically-driven neurorehabilitation study. In Aim 1, we will compare the effectiveness of short-duration intensive arm speed training with dose- equivalent accuracy training in individuals with mild to moderate impairments due to chronic supratentorial stroke. We hypothesize that, compared to accuracy training, speed training will improve speed, smoothness, and accuracy of arm reach movements, as well as overall arm function for up to 1-month post-training. In Aim 2, we will demonstrate, with the same participants, that speed training leads to improvements in anticipatory feedforward control. Using a fast elbow movement task, we will test the hypothesis that speed training improves both scaling of the elbow torques to reach different targets and anticipatory control, as measured by shoulder muscle activity that compensates for interaction torques at the shoulder. In Aim 3, we will determine the involvement of cerebellar-cortical circuits in the recovery of arm movements due to speed training. Using diffusion tensor imaging acquired before training for Aim 1, we will test the hypothesis that the extent of damage to cortico-cerebellar tracts predicts improvements in both reach movement parameters and feedforward control achieved through arm training.