Due to the high cost of motor bearing fatigue experiment and insufficient fault data, the effect of using artificial intelligence algorithms such as machine learning for fault diagnosis is not good. In addition, the accuracy of a single model for motor bearing fault diagnosis is also low. In order to solve these two problems, a fault diagnosis method for motor bearing fault diagnosis is proposed, which combines auxiliary classifier generative adversarial network (ACGAN) with model fusion. Firstly, the vibration data collected are converted into two-dimensional grayscale images, and each grayscale image is labeled and input into ACGAN model to generate a large number of new samples that highly fit the original data. Then the new samples are mixed with the original samples, and the data are dimensionality reduced to input into the model which is a fusion of six base-learners and one meta-learner. Finally, the diagnosis results are output by the fusion model. The test proves that ACGAN and model fusion can effectively improve the accuracy of fault diagnosis of motor bearing.