The dual-tree complex wavelet transform（DT-CWT） has the ability to characterize local feature of signal in both time-domain and frequency-domain

and the dual-tree complex wavelet has such merits as approximate shift invariance

good directional selectivity

and high computational efficiency.The information entropy based on wavelet can reveal statistic distribution properties of signal and find transient abnormal signal in system signal

so that some faults can be found early.In this paper

the partial discharge（PD） pulse waveforms which are generated by 4 typical insulation defects are transformed by DT-CWT

and then the energy and wavelet energy entropy in every coefficient are extracted as the features for pattern recognition.Discharge samples are obtained through large number of experiments

and a BP neural network

which plays the role of classifier

is established to recognize the PD signals generated by 4 typical insulation defects.The results show that the PD signals can be easily and sufficiently recognized by the neural network method using the features.