There is plenty of complex ground information in high-resolution remote sensing images. The direct road sementation in the images causes low accuracy and cannot rule out inferences such as residential areas. A road extraction method based on saliency analysis for high-resolution remote sensing images is proposed. The feature map of residential areas and roads is obtained by automatic seymentation. A saliency map of residential areas is obtained using the human visual system. The feature map of residential areas is generated by segmenting the saliency map. Finally, the roads are extracted by the logical exclusion OR operation of the two feature maps. Experimental results show that the proposed method can remove the inference residential areas effectively and extract roads perfectly. It has both theoretical and practical significance for road extraction in remote sensing images in the future.