The reason of the fault of threelevel inverter was analyzed, the open circuit fault of two IGBTs was classified, and all kinds of faults were coded. Taking the phase voltage signals of three bridge arms of the inverter circuit as the research object, and the threephase voltage signals were converted into two phases by using the d-q transformation to reduce the dimension of the fault information, and the feature vectors under different faults were extracted by Fourier transform. A fourlayer BP neural network was established and fault diagnosis was carried out. The network was trained and simulated with the feature vector as input, and the corresponding fault code as output. The feasibility and accuracy of fault diagnosis of threelevel inverter using BP neural network were realized and verified.