Based on deep belief network (DBN), a deep belief network classifier (DBNC) model is built. Analysis and typical data set validate the classification performance of it. Combining power transformer dissolved gas-in-oil analysis (DGA) data, a new transformer fault classification method, which is based on DBNC, is proposed initially. The approach uses the results of DGA as the necessary attributes to classify power transformer’s fault types. The proposed method is tested, which shows that it is suitable for power transformer fault classification. It has strong ability to extract features from samples, and has error-tolerance capability. Test results also show that the performance of the proposed approach prevails that of back propagation neural network (BPNN) and support vector machine (SVM) methods.