Precise knowledge of node location in wireless sensor network yields a wide variety of profound advantages. In this paper, localization scheme base on neural network for wireless sensor networks is described. We adopt neural network to model sensor network. The trained neural network can integrate global information of the sensor network as anchor nodes character, nodes deposited, environment, etc. The results of simulation demonstrate that the location scheme can obtain higher accuracy of location estimation, requires less anchor nodes and do not affected by NLOS environment and irregularity of anchor nodes radio power.