In order to restrain the end effect of EMD(empirical mode decomposition), a novel method that combines SVM (support vector machine) extension and window function is proposed. The combination makes up for the disadvantages that the SVM continuation fails to find the endpoint and window function could changes the original signal. First, the signal is extended at each extreme point by SVM and process the original data with a rectangular window and the continuation data with a hamming window. Second, employ EMD method to decompose the data and get a number of intrinsic mode functions(IMFs). Finally, the ends of the decomposed IMF component continuation section are removed in order to suppress the endpoint divergence. Orthogonality is chosen as the quantitative evaluation index. Simulation and experiments show that this method can effectively restrain the end effects during the process of empirical mode decomposition.