An algorithm for extracting road network by fusion of multi-band remote sensing images is presented in this paper. Firstly, straight lines and parallel lines are extracted from multi-band images and fused in order to overcome the uncertainty of the description of roads in the images. Secondly, the roads that fit well with the constraints of the model of road are recognized according to the local property of lines. Eventually, the roads that can not be fit well with the constraints of the model of road are recognized by suing the global connection constraints of road network. The algorithm is applied on 3-band remote sensing images, and the effectiveness is shown by the results.