Rolling bearings, as core components of rotating machinery, require timely and accurate identification of fault types to ensure the safe and stable operation of mechanical equipment. However, existing fault diagnosis techniques for rolling bearings face challenges in terms of comprehensive feature extraction and noise resistance. To address these issues, this paper proposes a fault diagnosis method based on a multi-scale convolutional neural network (MSCNN) integrated with a selective kernel attention mechanism. First, a MSCNN is constructed as the feature learning framework. It employs dilated convolutional kernels to capture fault features across multiple frequencies and incorporates batch normalization layers to mitigate overfitting. Second, the selective kernel attention mechanism is enhanced with adaptive max pooling, enabling the dynamic adjustment of pooling region sizes to effectively reduce information redundancy. Experimental validation using two bearing datasets demonstrates that the proposed method outperforms existing techniques in terms of fault diagnosis accuracy, robustness, and generalization capability.