Abstract: The dual-tree complex wavelet transform (DT-CWT) is adopted to make a multi-scale decomposition of UHF partial discharge (PD) signals, and an optimal algorithm for solving DT-CWT decomposition is proposed. In addition, the optimal complex wavelet energy (OCWE) features are extracted from the high-layer real and imaginary parts of UHF PD signals after decomposed by DT-CWT, and the fisher linear discriminant method is adopted to select the energy features. Finally, the selected features are used for PD type recognition. The results show that the high-layer wavelet energy features can effectively recognize four typical insulation defects in GIS with a recognition accuracy reaching 94.5% or above. It is proved that the OCWE features are more suitable for PD recognition.