In this paper, the wavelet threshold method was applied to ECG signal denoising. We proposed an improved threshold function on the basis of analysis and understanding of the commonly used ones. And the MIT-BIH database was used to simulate the ECG data in Matlab, respectively using common threshold functions and improved one for hard and soft thresholding to remove the noise. Experimental results show that soft thresholding method of improved threshold function can effectively filter out the main interference, and better to retain the characteristics of the ECG.