In order to improve the ability of monitor for winding deformation of transformer by frequency response analysis method , windowed calculation was proposed in this paper for improving traditional correlation coefficient method and diagnosed the fault type combined with fisher discriminant analysis. Firstly, a window with specified length is moved from the beginning to the end of frequency band of transfer function and scans the total frequency range of the transfer function with a desired step. At each step, the correlation coefficient are calculated at the window length frequency region and the calculated indices at each step can be used as a feature of frequency response curve after winding deformation. At last, because of the ability of the Fisher Discriminant Analysis for maximizing the between class separability while minimizing their within-class variability, Fisher Discriminant Analysis was utilized for dimension reduction from a lot of calculated features by the proposed method and pattern recognition for winding deformation.