This paper presents a new approach to detection of high impedance faults in distribution systems using artificial neural networks. The proposed neural network was trained by data from simulation of a distribution system under different fault conditions, and tested by data with different conditions. Details of the design procedure and the results of performance studies with the proposed method are given in the paper. Performance studies results show that the proposed algorithm is very good performance in detecting a high impedance fault with nonlinear arcing resistance. It is clearly shown that with this integrated approach, the accuracy in detection fault is significantly improved over other techniques based on a conventional algorithm.