Diseases or accidents often lead to hand dysfunction. At present, the assessment of hand rehabilitation mainly depends on the experience and judgment of doctors. Most of hand rehabilitation products on the market are complex and expensive. The daily training data for further diagnosis of patients cannot be recorded by the existed products. In this paper, a data glove with multisensors is designed for hand rehabilitation. The least square approach is adopted to correct the accelerometer and magnetometer. The complementary filter algorithm is used for sensor data fusion. Realtime hand gesture can be obtained and demonstrated in Unity 3D platform. Four kinds of gestures are captured and reconstructed by the proposed approach. It reveals that the data glove can effectively improve the hand rehabilitation procedure for both patients and doctors.