The algorithm of GA optimized BP neural network is introduced into the line selection method of small current grounding fault. This paper conducts simulation experiment based on MATLAB, extracts the various characteristic quantities of zero-sequence current signal by traditional line selection methods , such as wavelet packet method, fifth harmonic method , fundamental wave ratio , amplitude ratio and phase ratio method, and calculates the fault measure data using fault measure function, and then, inputs the data into GA-BP neutral network and single BP neutral network respectively for training and testing, discusses the difference in line selection performance between GA-BP neural network algorithm and single BP neural network algorithm, and outputs the fault line selection results and compares with the fault measure data based on each line selection method. The results show that the GA- BP neutral network combining multiple traditional line selection methods has much higher accuracy rate than transitional counterparts, and its selection speed and precision is superior over single BP neutral network, which can carry out fault line selection in a quick and effective way , satisfying the requirements of fault line selection in the distribution network.