Learning a motor task in one condition typically generalizes to another, although it is unclear why it generalizes substantially in certain situations (e.g., across limb configurations and workspaces), but only partially in other situations (e.g., across movement directions and motor effectors). Previous accounts of motor generalization often focused on the idea of error-based learning. However, this idea cannot explain clearly why motor learning would generalize substantially in certain situations, but only partially in other situations. Through a series of studies, we show that motor generalization is limited when it only relies on error-based learning, and also that the extent of motor generalization can be enhanced substantially, not only across movement directions, but also across motor effectors, when it involves both error-based and use-dependent learning. These findings have implications for rehabilitation that involves unilateral limb impairment.