Radar signals are of non-stationary, so extracting effective classification features from them by general transformation methods is very difficult. Based on the characters of wavelet packet bases, a method of recognizing radar target is presented in this paper. The target recognition observes the following process - first resolving the wavelet of radar echo-signal, selecting the optimum wavelet bases via given training sample sets based on distance criteria to get the feature vector for target recognition, thus the targets are recognized by the effective features. The experiments of recognition using the data of three kinds of aircraft models are performed and the result demonstrates high recognition rate and low sensitivity to noise by using this method.