Power transformer is one of the important equipment in power system operation, correct diagnosis of the fault and defects is related to the safe operation of the entire grid. Support vector machine(SVM) can better solve the multi-classification with small sample and nonlinear characteristics, it is suitable for fault diagnosis of transformer. In this paper, we get the best global optimal solution of SVM using the cuckoo search algorithm, and get the SVM classification model with the best parameters. In this classification model, the relative content of each gas of dissolved gas analysis (DGA) is put as the evaluation indexes. The transformer fault is divided into 4 types of fault, i.e. low energy discharge, high energy discharge, mid-low temperature overheating, and high temperature overheating. Through the analysis of the existing data instance, the accuracy of the classification model using cuckoo search algorithm is better than that using grid search (GS), particle swarm optimization (PSO) and genetic algorithm (GA).