High impedance faults due to the high magnitude of impedance in their paths have little fault current. Therefore, network security devices are generally unable to identify them. This paper presents a new method for identifying high impedance faults. First, the high impedance waveforms that have been extracted from actual experiments are de-noising by the wavelet transform. The waveform of other transient network conditions is also derived from the simulation of the standard 13-bay IEEE distribution system in PSCAD software. Then, by using time-frequency transformation S, extraction of the selected features of this waveform is performed by principle  components  analysis(PCA) of the monitoring sample points and entering the neural network. Finally, to distinction between transient states and high impedance faults and detection this fault, the multilayer perceptron neural network has been used. The results show clearly the effectiveness of this method