Transformer fault diagnosis is an important technical means to ensure the safety operation of power system. In order to improve the accuracy of fault diagnosis of transformer, this paper proposes an method which based on LSTSVM optimized by bat algorithm. For the multiple classification problem of transformer fault diagnosis, in order to reduce the accumulation of errors and improve the accuracy, a huffman tree is build according to dissimilarity matrix between class, then we established a multi-classification fault diagnosis model based on LSTSVM. And parameter of each classifier in the model is optimized by bat algorithm. The simulation results show that the method in this paper can achieve higher accuracy compared to other method.