In order to enhance effectiveness of single-phase ground fault feature extraction and to achieve exactly identification of fault line selection，a new fault line selection method of single phase grounding fault based on wavelet packet and online sequential extreme learning machine ( OS-ELM) is proposed. In this paper，the fault characteristics of fault transient signals are extracted by using the multi-resolution frequency division characteristics of wavelet packets. Combining batch and successive iterative，and constantly updating the training data and network structure，to achieve quickly fault line selection of single-phase grounding. The simulation results show that the more effective fault characteristics can be obtained using wavelet packet full frequency analysis，and OS-ELM is superior to the back propagation ( BP) network，radial basis function network ( ＲBF) and support vector machine ( SVM) in test time and diagnostic accuracy. The proposed method is effective in the single phase grounding fault line，can better meet the needs of online application.