Abstract: It is hard to improve the transformer fault diagnosis accuracy only based on single feature information.While multi-kernel learning relevance vector machine（MKL-RVM） can enable informative integration of possibly heterogeneous sources.A new power transformer fault diagnosis method based on MKL-RVM was proposed.The method integrated the feature information of power transformer operating state,and outputted the probabilities of various power transformer operating states,giving more available information for the maintenance and repair of the power transformer.Additionally,in order to enhance the performance of the MKL-RVM,genetic algorithm combined with K-cross validation was adopted to optimize the kernel function parameters.Experimental results showed that the proposed method was capable of more excellent diagnosis accuracy to back propagation neural network（BPNN） and support vector machine（SVM）.