In the actual seismic record formation process, there is a superposition of high and low frequency components of seismic waves. The composition of high and low frequency components of ground motion is an important reflection of the characteristics of the response spectrum of ground motion. This study is based on the digital filtering technology, taking the average period of acceleration response spectrum (T_avg) 2 s as the cut-off period and combining with empirical mode decomposition(EMD), and it extracted the intrinsic mode function(IMF) and residual component which the cut-off period above 2 seconds as long-period components of long period ground motions(LPGM) after refactoring and baseline correction. The remaining IMFs were reorganized into short-period components; The validity of the extraction method is verified by correlation between seismic records and their components and the discreteness of the original ground motion and the response spectrum. It also compare this method with existing methods. The results show that the extraction method for LPGM components in this paper can better extract the long and short period components of the seismic response spectrum that are less discrete and have a good correlation with the original ground motions in the long and short periods respectively. It better reflects the spectral characteristics of original ground motions and has better adaptability in the LPGM component separation.