To deal with the issue of fault diagnosis of multilevel inverter, a deep neural network based approach is proposed for the fault diagnosis of multilevel cascade H-bridge inverter. Firstly, the fault model of multilevel cascade H-bridge inverter is introduced. Then, the fault features are extracted directly from the original fault signals by using the stacked autoencoder based deep neural network. Finally, the SOFTMAX classifier is used for the classification of the fault features to carry out the fault detection and diagnosis of multilevel inverter. Several simulation experiments are implemented based on MATLAB/Simulink to test the performance of the proposed approach, which is superior to other traditional intelligent fault diagnosis algorithms on both accuracy and robustness.