Due to the declining accuracy caused by noise in wireless sensor networks, a data fusion method called normal score test, based on sliding window, is proposed. With this method, a measured data sample in a cycle is taken as a sliding window, and a double sample normal score test is employed into samples. Furthermore, a relationship matrix is created to maximize the adjacent subgraph. An estimated fusion value is achieved in terms of weighting and averaging its vertices. To some extent, this method makes good use of the character of a normal score test and restrains the impact noise on fusion. Simulation results demonstrate the effectiveness of proposed method.