Aiming at the problem of insufficient feature extraction and low recognition rate of convolutional neural network, a novel facial expression recognition method based on multi-feature fusion convolutional neural network is proposed. First, to increase the width and depth of the network, Inception architecture is introduced into the network to extract the diversity of features; Then, the extracted high-level features are fused with the low-level features, and the pooled features are used to send the fused features into the full connection layer, then the fused features are processed to increase the non-linear expression of the network and enrich the features learned by the network. Finally, the output layer classifies the expressions by Softmax classifier, conductes experiments on FER2013 and CK+, and analyzes the experimental results. Experimental results show that the improved network structure improves the recognition rate of facial expressions in FER2013 and CK+ data sets by 006% and 225%, respectively. The proposed method is valuable for setting up convolution neural network and parameter configuration in facial expression recognition.