As a core rotating component in printing press, the operation status of bearing plays a major role in the health monitoring of printing press. The work aims to propose an intelligent diagnosis method of bearing faults in printing press by mixing the advantages of wavelet time-frequency processing with the Inception v3 model. The Morlet wavelet was used to process the raw vibration signals collected from bearing, and the corresponding two-dimensional time-frequency images were obtained to characterize the bearings faults from the time-domain and frequency-domain. The time-frequency images were used as input of the Inception v3 model, and the filter-level sparsity of the Inception v3 model was used to quickly and automatically learn the fault features from the time-frequency images and adjust the model parameters; finally, the trained model was used to implement the fault diagnosis of printing press bearing. The effectiveness of the proposed method was verified with a printing press experimental platform, and the results indicated that the average diagnostic accuracy of the method can reach 92.53%. Compared with traditional intelligent diagnosis methods, the proposed method has higher diagnosis accuracy and stability to achieve the bearings fault diagnosis of high-precision printing press.