In order to improve the precision of single phase to earth fault line selection device for the small current grounding system, an improved fault line detection method using RBF neural network based on crisscross optimization algorithm (CSO) optimization is presented. By means of Matlab/Simulink, numerous single-phase-to-earth fault experiments are carried out and many zero sequence current signals are obtained. The fault characteristics of the zero sequence current signals, such as transient characteristic value, active component and five harmonic component, are extracted by wavelet packet transform and Fourier transform. Then choosing the extracted characteristic quantity as the input of neural network, using neural network which has been optimized by CSO to train fault characteristic value, the fault line selection can be realized. Through the simulation, 100 sets of different fault samples are set up, and the 80 groups are set as training sets, and the 20 group is the test set. The simulation results indicate that compared with the traditional neural network, the proposed method has better training effect and higher precision.