In view of the traditional SVM method may lead to false diagnosis when the sample fault feature is unclear, a fault diagnosis method is proposed based on the multi - classified probability output model. The multi - classified probability output model is constructed by using Sigmoid function based on SVM. The input of the model is the DGA data and the fault type of transformer, the output is the probability estimate for each failure type. The fault probability information can be used to identify whether the fault feature is clear by setting the relevant evaluation standards. The example shows that the samples with unknown fault characteristics can be identified by using the diagnostic results of the MCPO model, and provide a reference for further reasonable measures.