The accuracy of current road extraction algorithms for high resolution remote sensing images is low. Aiming at this problem, a road extraction algorithm for remote sensing images based on the improved expectation-maximization (EM) clustering is proposed. First, the morphological preprocessing is carried out to remove the interference information from the road. Then, the improved EM clustering is applied to determine the number of segmentation regions and segment the images automatically. And the extraction of roads from remote sensing images is finally completed through the contour detection and gray histogram thresholding. Experimental results show that the proposed algorithm can effectively screen the noise on the road and extract the main road information, with high accuracy of over 90% and low redundancy.