According to the principal component analysis (PCA) and linear discriminant analysis (LDA) identification accuracy is not high in the ear recognition process. Ear recognition algorithm based on discrete wavelet transform (DWT), PCA and LDA is proposed. The algorithm is decompose the human ear image with two-dimensional DWT, select low-frequency sub-band that contains the most image information, use PCA and LDA in succession to extract the optimal sample mapping space, and use the nearest neighbor rule to classify the human ear image. The experimental results show that this method is superior to identification method of PCA and LDA.