A facial expression recognition method is proposed. The Gabor features of facial expressional images are extracted and twice feature selection is used to recognize facial expression. Since the high-dimensional Gabor features are quite redundant, the high-dimensional features are sampled and the 2-D PCA is used to select valid features. Then, seven expressions are recognized based on multi-classifier combination with fuzzy integral. The experimental results on JAFFE facial expression database show that the proposed method is valid. It has higher computational speed and better recognition rate than 2-D PCA.