In this study, a new bridge weigh-in-motion (BWIM) algorithm based on the convolutional neural net? work technology was developed for identifying the axle weights of multi-vehicles crossing the bridge. First of all, the bridge strains under vehicular loading with variable weight were simulated using the vehicle-bridge coupling simula? tion system. Then, a convolutional neural network, consisted of nine convolutional layers and two fully connected lay? ers, was developed based on an open source framework for deep learning, i.e., KERAS. The convolutional neural net? work was trained by Adam optimizer to map the relationship between the bridge strain and the vehicle weight under different scenarios, and optimized by minimizing the fitting error. Eventually, the identification accuracy of the pro? posed BWIM system was analyzed under the conditions of single- and multi-vehicle loadings. The results show that the mean identifying error of the proposed BWIM was less than 5% for both the single- and multi-vehicle scenarios, and changed slightly with the varying traveling speeds and lateral loading positions, indicating the good and stable performance of the proposed BWIM algorithm in axle weight identification. In addition, the proposed BWIM system doesn’t need a bridge influence line in advance to identify the axle weight, and therefore provides an alternative technology for bridges that are not suitable for the influence line method.