Face expression recognition is an important branch of image processing and a hot topic in computer vision research. Thetraditional convolutional neural network improve the accuracy of the network by deepening the number of network layers and ex-panding the scale of model parameters, but compared with the traditional machine learning algorithm, its accuracy has not been sig-nificantly improved. For this reason, inspired by DenseNet, a network model M-Densenet is designed, which is specifically appliedto facial expression recognition. This model is used to conduct experiments on Fer2013gray expression recognition data set. The ex-perimental results show that the model is used to conduct experiments on Fer2013gray expression recognition data set. With a highaccuracy rate of70.45%, compared with the traditional convolutional neural network, the number of parameters of the model is low-er and the utilization rate of model parameters is higher.