Abstract: In order to avoid the complex manual feature extraction process in traditional facial expression recognition methods and to extract more facial features, we propose a Dual-Path Feature Fusion model that combines Convolutional Neural Network (CNN) with Histogram of Oriented Gradient (HOG).In the first channel, the facial expression image is normalized, and the trainable convolution kernel is used to extract the implicit features.In the second channel, the HOG of facial expression are extracted and transmitted to the full connection layer of CNN.Finally, the fusion features are transferred to the output layer, and the recognition results are obtained via the Softmax classifier.This paper conducted an experiment on FER2013 & CK+ database, the results verify that the method presented is effective.