Abstract: Partial discharge (PD) pattern recognition is a critical indicator for evaluating the insulation state of electrical equipment. PD signals can be obtained by ultra high frequency(UHF) sensor. However, traditional methods based on statistical parameters have the disadvantages of high dimension and invalid information for signal feature extraction. In this paper, we propose a feature extraction method based on time-frequency analysis and fractal theory for partial discharge pattern recognition of gas insulated switchgear (GIS). Firstly, the wavelet transform is applied to extract the energy map of time-frequency distribution of partial discharge signals. Then, the differential box counting (DBC) is used to extract the fractal dimensions features of the energy distribution map, and the linear discriminant analysis (LDA) is used to reduce the dimension of the feature vectors. Finally, the support vector machine(SVM) is employed to classify the partial discharge defects. To evaluate the effectiveness of the proposed method in this paper, four typical types of defects, namely protrusion discharge, free moving particle discharge, surface discharge, and floating electrode discharge, have been set up in the 252 kV GIS model chamber of the laboratory. Partial discharge signals have been obtained by UHF sensors and classified by the proposed method. The experimental results show that the recognition accuracy rate for all the four typical discharge types exceeds 96% by using the proposed feature extraction method, which is significantly better than the traditional methods based on statistical parameters for signal feature extraction.