Gender is the most basic information on human beings. It is of great significance in the field of face recognition. However, due to factors such as lighting, poses, and expressions, gender classification at this stage has the problems of low accuracy and slow training speed. In this paper, various variants of convolutional neural networks are used to deal with gender classification tasks, namely VGG16, Inception V3 and ResNet50. Besides, the whole process from dataset selection, dataset processing to face gender classification is described. The experimental results show that face gender classification based on convolutional neural networks can improve training speed and classification accuracy, and the highest classification accuracy can reach 95.10%.