Abstract: Using empirical mode decomposition (EMD) for denoising, it's difficult to remove the noise from the signal because of the mode mixing in EMD. Aiming at this problem, a new method was proposed that is extrema-mean empirical mode decomposition (Extrema-mean EMD), making a new definition of intrinsic mode function（IMF). Firstly,the average of each two successive extrema was calculated and the mean curve was obtained. Secondly, the mean curve was subtracted from the signal iteratively, and then the IMF was got. Finally, the high frequence IMF were removed from the original signal for denoising. Through the random signal simulation and lidar return signals de-noising experiment, it's turned out that compared with the EMD, the Extrema-mean EMD could restrain the mode mixing and remove the noise from the signal effectively. This method can decrease the MSE of the denoised signal significantly. Therefore, the Extrema-mean EMD has a promising future.