An analysis of the near surface atmospheric electric field data in the summer of 2009 is presented based on the empirical mode decomposition (EMD) method. The vari scaled components of the atmospheric electric field in thunderstorm and fair weather are decomposed and the atmospheric electric field oscillation characteristics of two types of weather conditions are extracted and compared. The results show that the EMD method is suitable for the analysis of atmospheric electric field data. The atmospheric electric field in thunderstorm weather is under the background of atmospheric electric field in fair weather, and so contains the steady periodic oscillation compositions of fair weather. The atmospheric electric field energy in fair weather is concentrated in the long period oscillation component, while that in thunderstorm weather is mainly concentrated in the short period oscillation component. Before cloud to ground lightning occurring, the central frequency of IMF1 (IMF: Intrinsic Mode Function) will jump or the corresponding amplitude of IMF1 will increased significantly. According to these characteristics, 38 thunderstorms selected randomly are tested with lightning location data and the results show that the detection probability of warning is 842%.