This paper presents a wearable hand movement rehabilitation system for stroke patients. The system is developed based on data glove and keyboard games. Rehabilitation practice is achieved via hand gesture recognition. In this work, the data glove with bending sensors is good for motion data collection during hand movement rehabilitation. The hand animation model, combined with keyboard games, enables the stroke patient under test to see its fingers movements and exercise process. In feedback stage, the rehabilitation evaluation and recommendation are provided based on the recognition of hand gestures. The experimental results have demonstrated a high accuracy on overt gesture recognition and a reasonable accuracy on complex key press gesture recognition.