The invention discloses a facial expression recognition method based on partial shielding of improved VGG-Net. The method comprises the following steps: reconstructing a facial expression recognitiondata set under partial occlusion; preprocessing an input image; performing improvement and model training based on VGG-16, connecting two residual blocks behind a 13th convolution layer, and obtainingan output result vector through a full connection layer and softmax loss, thereby realizing image processing and weight learning. According to the method, the traditional softmax loss is improved into a Large Margin Cosine Loss (LMCL), and an output result vector is obtained. Compared with the latest research method, the partially shielded facial expression recognition rate of the method has theadvantages that on one hand, the accuracy of the method is as high as 3% compared with that of other methods, and on the other hand, the experimental data result of the method provides data which canbe referenced in the research direction.