In order to improve the practicability of face recognition technology, a new face recognition method was proposed by adopting the facial mirror symmetry and Kernel Principle Component Analysis (KPCA). Firstly, the original images were decomposed by wavelet transform, and the low frequency components could be obtained. Then, the odd symmetry samples and the even symmetry samples were obtained by mirror transforming. Odd/even eigen vector were separately extracted through KPCA and fused to composite features by an odd-even weighted factor. A nearest neighbor classifier was used to classify the images. The proposed method was tested on the ORL face image database. The experimental results show the method can increase the sample capacity, overcome the effect of illumination and posture, and raise the recognition rate. Besides, in the comprehensive performance, it is better than contrast method.