This paper proposes a novel facial expression recognition method by extracting the wavelet moment invariants of the images as feature vectors, and using AdaBoost to select effective features. Wavelet moment invariants can present the facial expressions effectively and invariant under translation, scaling and rotation. To reduce the dimensions and eliminate the redundancy of feature vectors, we utilize modified AdaBoost algorithm to select the combination of the effective features that best classify the samples. Experimental results indicate that the proposed method outperforms conventional methods, such as Gabor and Zernike moments.