The electric machine has been widely used in various fields, and its failure may not only cause damage to the machine, but also many problems, such as economic loss, casualties and so on. Therefore, it is important to apply timely and efficient fault diagnosis technology. Deep learning has been applied in fault diagnosis of electric machines and obtained some useful results because of its more powerful and more complex feature expression ability than traditional techniques. Hence, this paper introduces four classic types of deep learning model, the Deep Belief Networks (DBN), Auto-Encoders (AE), Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN), and summarizes the application of these four deep learning models in electric machine fault diagnosis. Finally, the problems and challenges that deep learning faces in this application are summarized and prospects discussed. This work is supported by National Natural Science Foundation of China (No. 51637001, No. 51607001, and No. 51507002) and Natural Science Foundation of Anhui Province (No. 1508085ME87 and No. 1708085QE108).