Learning to move the hands in particular ways is essential in many training andleisure virtual reality applications, yet challenging. Existing techniques that support learning of motor movement in virtual reality rely on external cues such as arrows showing where to move or transparent hands showing the target movement. We propose a technique where the avatar's hand movement is corrected to be closer to the target movement. This embeds guidance in the user's avatar, instead of in external cues and minimizes visual distraction. Through two experiments, we found that such movement guidance improves the short-term retention of the target movement when compared to a control condition without guidance.