In order to obtain the effectual feature of signals, wavelet packet transform is used. The characters of every wavelet packet basis are different, which can express the main feature of a signal. The local discriminant basis (LDB) is calculated based on the distance criterion, and a feature extraction method is proposed. The feature vector, which expresses the energy of sub-space in LDB, is obtained by using Fisher criterion for feature choice. The classification experiment for three different classes of targets is done. The results of the experiment show that this feature extraction method is effectual in pattern recognition.