Aiming at the problem of opencircuit fault arising in diodeclamped threelevel inverter, a new fault diagnosis method based on decision tree support vector machine (DTSVM) is proposed. Taking the inverter state as an example, firstly, the operation conditions of main circuit in inverter are analyzed to classify faults. Then, in terms of the multiscale decomposition of wavelet analysis, the middle, upper and down bridge voltages are selected to extract the fault features, respectively. Moreover, particle swarm clustering algorithm is built to construct the DTSVM classify model, and the multimodel fault diagnosis of power component in threelevel inverter is finally accomplished. The simulation results show that this method in case of less classification model to complete fault diagnosis, comparing to other methods such as back propagation neural network, oneversusone support vector machine and extreme learning machine, the diagnostic accuracy up to 98.46% for multimode fault diagnosis of threelevel inverter in 10% white noise, which indicate that the algorithm has better accuracy and robustness.