Rehabilitation programs designed to develop skill in upper extremity (UE) function after stroke require learnercentered opportunities for active problem solving. Virtual realty (VR) provides a unique environment where the presentation of stimuli can be systematically controlled to enable an optimal level of challenge by progressing task difficulty as performance improves. We describe four VR tasks that were developed and tested to improve skilled arm and hand movements in individuals with hemiparesis. Two participants post-stroke with different levels of motor severity attended 12 training sessions lasting 1 to 2 hours each over a 3-week period. Behavioral measures and questionnaires were administered pre-, mid-, and post-training. The less impaired participant averaged more time on task, practiced a greater number of blocks per session, and progressed at a faster rate over sessions than the more impaired participant. Differences in functional outcomes for these two cases can be explained in part by which tasks were practiced, the level of task difficulty applied during practice, and the amount of repetition included in practice.