The traditional facial expression recognition (FER) methods only extract single expression feature. Meanwhile, the choice of expression classifiers has limitations. To solve these problems, we propose a FER method based on the fusion of local features and deep belief network (DBN). Firstly, the eyebrows and eyes part and mouth part with rich expression information are extracted as local expression images. In order to attain more effective expression features, the Log-Gabor features with texture information and second-order histogram of gradient direction features with shape information are extracted and fused from local expression images. DBN model is trained with fusion features. The trained DBN model is used to recognize the facial expression. The experimental results show that the recognition rates of the proposed method on three databases are 96.30%, 97.39% and 95.73%. The proposed method effectively improves the recognition rate of facial expression.