In order to enhance the safety, reliability and maintainability of automotive fuel cell system, considering that the large quantities of complete faults sample data are difficult to obtain, a fault diagnosis method of support vector machine(SVM) based on binary tree multi-classifier is put forward. Firstly, based on the 60kW automotive fuel cell system designed by our group, the fault mechanism and characteristic are analyzed. Then, 15 kinds of fused fault scenario are treated as the inputs of SVM after normalization, and 14 kinds of typical faults are taken as the outputs of SVM, in the training process, 310 groups of sample data and the RBF kernel function are adopted, and the penalty parameter as well as kernal parameter of SVM are optimized with particle swarm optimization(PSO) algorithm, all the different typical faults can be identified from the experimental results with 90 groups of testing data. Finally, the fault diagnosis performance of SVM is compared with that of BP neural networks with different quantities of training sample data, the simulation results demonstrate that the proposed method has better capability of faults classification and generalization, and can be effectively implemented in the fault diagnosis of automotive fuel cell system.