After single phase-to-ground fault occurs in asymmetrical power network, statistical characteristics of the voltage data distribution between the fault phase and the non-fault phase are significantly different. So a novel method for fault phase selection based on frequency distribution theory is proposed. Firstly, it selects per phase voltage data in the first 1/2 power frequency cycle after the fault occurred, and the extended prony algorithm is used to extract the power frequency components of each phase voltage to get the range of class interval. Then, these sampling data are split up into several groups by means of the frequency distribution theory, and frequency distribution values of per phase voltage are obtained by Bessel formula. Finally, the fault phase can be detected directly by comparing the magnitude of it. Matlab simulation results and adaptability analysis verify that the proposed method has high accuracy and strong adaptability. This work is supported by Natural Science Foundation of Shandong Province (No. ZR2016EEM13).