Attenuating seismic noise covering the reflected signal components is an important step in seismic data conditioning. There are several methods to improve the quality of seismic data and to increase the signal to noise ratio, which empirical mode decomposition is an effective one. In this study, empirical mode decomposition, ensemble empirical mode decomposition, complete ensemble empirical mode decomposition methods and their application in random noise attenuation have been investigated. In these three methods, the noisy signal is decomposed into several sub-signals consisting specific amplitudes and frequencies called intrinsic mode functions (IMFs) with frequency bandwidth descending orders. The IMFs containing random noise are ignored and remained IMFs are combined to recover the original seismic signal. The results show that the complete ensemble empirical mode decomposition method has a better performance in random noise attenuation than the two other methods.