The wheel thread flat fault signals is non-stationary signals.To identify the wheel thread flat,a new fault diagnosis method was put forward,which combined the Empirical Mode Decomposition(EMD)with Artificial Neural Networks(ANNs).The wheel thread flat fault signals were collected by experimental tests and were analyzed using EMD to get the Intrinsic Mode Function(IMF)components.The energy and kurtosis features of these IMF components were extracted to construct the feature vector which was served as input parameters of the neural network to identify the fault pattern of the wheel/rail system.This identification procedure was implemented with LabVIEW and was verified by experiments.The experimental result shows that this method can effectively identify the wheel tread flat fault in the wheelset speed range of 0~200 km/h,and has a wider range of application.