Most of high resistance faults in distribution network are single-phase earth faults caused by overhead line contacting high impedance objects. These faults are difficult to identify by steady-state features. In this paper, the high frequency components of faults are analyzed based on equivalent network and the influence of normal state on identification are considered. An identification method of single-phase high resistance earth fault in distribution network based on wavelet packet energy ratio of zero sequence voltage is then proposed. Wavelet packet decomposition is used to extract energy features of zero sequence voltage. The ratio of energy in low-frequency band to energy in high-frequency band is used as the identification feature of high resistance faults. Based on a typical radial 10 kV distribution network, the performance of the proposed identification feature is evaluated. The validity of the method is evaluated by the identification and comparison results of BP neural network. The applicability of the method is verified by the analysis and simulation of different neutral grounding modes, different topologies and the change of fault conditions.