The method of extreme point extension based on clustering least squares and slope was proposed to address the endpoint effect that could occur in signal denoising by empirical mode decomposition (EMD). Fully considering the impact of noise on the extension method, the clustering least squares method was used to describe the trend of global extreme points, and the coordinate information of the extension points was determined by combining the local characteristics of the signal boundary. The performance of the algorithm was evaluated by similarity coefficient, root mean square error and orthogonality level through the study of simulation signal and gyro example signal. The experimental results show that the method proposed can effectively suppress the endpoint effect of EMD and avoid the pollution of the endpoint effect to the intermediate data to the greatest extent, thus improving the accuracy and reliability of signal denoising, and providing a feasible and effective solution for the application of EMD in signal processing.