Abstract: In recent years,empirical mode decomposition (EMD) has gotten a lot of public attention due to its capability in processing non-stationary seismic signal and its convenience to implement.In this paper,we summarized the existing applications of EMD in seismic data denoising,and proposed a new approach combining EMD in f-x domain and multi-channel singular spectrum analysis (MSSA) to attenuate random noise.Comparing those approaches combining f-x EMD with other methods such as prediction filtering in f-x domain,wavelet threshold and curvelet transformation,the new approach can obtain higher SNR compared to f-x MSSA and can predict the lost linear energy due to f-x EMD.The workflow of the method is as follows:implementing f-x EMD to the given seismic data at first,all the relative horizontal events can be retained,leaving part of useful dipping signals and some random noises in the difference profile;then f-x MSSA can be used in the difference profile to restore dipping signals;finally resultful profile can be gotlen by horizontal signals plus dipping signals.Two synthetic and a real data examples demonstrate the superiority of the proposed approach.