A definition of wavelet entropy weight is proposed on the basis of both the result of transient current signal wavelet analysis and the idea of entropy weight. By using this definition and high frequency transient signal， a fault phase selecting method is represented. This method takes the transient high-frequency current of current transformer as its signal， and uses wavelet to pick up signal’s feature. Calculating this feature by its scale distribution weight， transient signal wavelet entropy can be obtained. Based on above work， a fault phase selecting criterion is presented. All faults of a typical 500 kV line under EMTDC and MATLAB by using the method are simulated. The result shows that this method is quick， accurate and free from the disturbance of transition resistor， fault time and location.