Bearing faults are one of the most common faults in industrial equipment, and their effective detection is crucial for extending the service life of the equipment. Research on bearing faults based on vibration signals has rapidly developed in recent years. Extracting representative features from the original signal completely and effectively is key to solving such problems, and existing methods have not fully addressed this issue. This paper proposes a bearing fault detection model based on Feature Mode Decomposition (FMD) and Transformer, which decomposes the original complex signal into several simple basic modal signals, enabling effective feature extraction. The ability of a transformer to handle long-distance information can effectively process one-dimensional bearing vibration signals. Experimental results show that the classification accuracy of FMD-Transformer is over 95% on the Ottawa bearing vibration dataset. Combining FMD and neural network models provides new ideas and directions for bearing fault research.