The method dealt with in this paper is very suitable for non-stationary nonlinear multi-component signal analysis. This paper discusses the basic principle of EMD, and studies a signal denoising method based on EMD. According to the different feature scales of EMD, the signal is decomposed into different frequency IMF components. Through deleting the high frequency IMF components full of noise and choosing the low frequency or the specified frequency IMF components, the signal can be reconstructed, thus achieving denoising. The results of simulation signal and measured data both show that the method can not only effectively remove the deterministic or random noise but also preserve the effective signal as much as possible and reduce the signal loss, thus improving the accuracy of data processing.