Aiming at the traditional rolling bearing fault diagnosis model feature extraction is relatively single, low accuracy, poor noise resistance, this paper proposes a bearing fault diagnosis method based on multi-scale convolutional neural network and Bayesian optimization of support vector machine (MS-CNN-BO-SVM), which enhances the original vibration signal with data, followed by the use of multi-scale convolutional neural network for feature extraction. Then, the extracted features are input into the support vector machine model optimized by Bayesian method to find the best support vector machine parameters by using the characteristics of Bayesian optimization for fast optimization. The experimental results show that the method proposed in this paper can better extract multi-scale fault features, has higher diagnostic accuracy compared with other similar fault diagnosis methods, and also maintains high noise immunity performance in different noise states, which can better solve the bearing fault diagnosis problem. This study is of great significance to the problem of motor bearing fault diagnosis.