The model recognition accuracy is low due to the scarcity of fault sample data in practical engineering. To address this issue, a rolling bearing fault diagnosis model based on the self-calibrated convolutional neural network ( SC-CNN) is proposed and applied to fault identification under the condition of small samples. Firstly, the BN algorithm is added after each convolutional layer to reduce the data distribution difference of different signals. Secondly, the self-calibrated convolution is adopted to learn the multi-scale features of the signal to improve the ability of the model to obtain useful fault features. Then, the channel self-attention mechanism is introduced to establish the correlation between channel feature information to highlight the fault features and suppress data overfitting. Further, a small number of training samples are fed into the model for learning. Finally, the fault signals under various conditions are taken as the input of the trained SC-CNN model for identification and classification. Evaluation experiments are implemented on two datasets. Results show that the recognition accuracy values of the proposed model are 98. 64% and 99. 83% under strong noise environment with SNR of -4 dB. Those two values are 94. 37% and 99. 64% under variable working conditions. Results show that the SC-CNN model has strong robustness and generalization performance under small sample condition.