Frequency response method was used to detect the power transformer winding deformation with the changing of peaks and troughs and frequency spectrum frequency in the curves of high middle lower spectrum. Some types of winding deformation was similar in the reflection of frequency curve. It would lead to the mistake results in the judgment of deformation by the reason of without the quantitative judgment. The paper came up with the SVM method in detecting the type of power transformer winding deformation, which can improve the recognition accuracy and efficiency of winding deformation. In order to set up the SVM model accurately of power transformer winding deformation recognition. Firstly the paper made choice of the characteristic, which could inflect the overall frequency response curve characteristic quantities. Secondly, the three types fault of the power transformer winding, including axis offsetting, picture to the deformation, winding changing between bread, chose the appropriate training samples and testing samples. Finally, the paper selected the optimal punishment factor and kernel function, in order to improve the accuracy of the SVM model for power transformer. The paper could get the three types fault identification accuracy of the power transformer winding with the identification of the test samples. The test results showed that the SVM model could dramatically eliminated the influence of the artificial misjudgment, and it realized the detection of transformer winding accurately.