The tracking based on Kalman filter fusion requires accurate system model and exact apriori information. Therefore, a novel method based on Kalman filter fusion in weighted covariance is proposed, which can increase the weighting factor of the sensor with less error according to the criterion of least squares. The Kalman filter fusion method can retain effectively valuable information and suppress noise. In such application as target tracking, even if the statistic information of noise is unknown, but correlative, optimal state estimation for target is still carried out by the proposed method in the criterion of least squares.