There is a problem that a single-channel fault classifier cannot fully express three-phase fault characteristic information and the classification accuracy is not high. Thus a transmission line fault classification method based on a multi-channel convolutional bidirectional long and short-term memory neural network (MCCNN-BiLSTM) is proposed. This method can input more than one fault three-phase signal at the same time, and can effectively extract the spatial and temporal characteristics of the fault signals, realize the comprehensive extraction of the three-phase fault signal features, and effectively improve the classification accuracy of the neural network. Based on a large amount of fault data analysis of the 735 kV three-phase series compensation transmission line model, no feature extraction algorithm is used for the three-phase fault voltage signal, and only the three-phase voltage amplitude data of the fault period is intercepted as the basic fault characteristic signal input. Simulations show that the network can quickly and accurately classify and identify 11 types of faults, and is not easily affected by factors such as the time of the fault nor excessive resistance. It has good robustness and adaptability. This work is supported by the National Natural Science Foundation of China (No. 51507015).