In the case of any prior knowledge is unknown, consistent data fusion algorithm is widely studied. Based on the existing methods, a novel limited memory fusion of multi-sensor based on support degree is proposed to avoid data saturation when the old measurement information increasing. The algorithm takes use of the exponential function to build the support degree matrix, and each sensor's consistency value of different time can be obtained. In the progress of iterative estimation, the limited memory method is used to just make full use of the last measurement information, and so the weight coefficient for sensor can be determined dynamically. Compare with the mean and support degree based data fusion techniques, the fusion weight assignment is more reasonable and the fusion precision can be improved further by the proposed method.