Abstract: Objective This paper presents a shape-based approach to extract arterial roads from urban road net-works is appropriate to the high level-of-detail found in volunteered geographic information(VGI)datasets.Firstly,the method calculates several shape descriptors of each polygon surrounded byroads.Secondly,some candidate arterial-road polygons are classified as seeds from all polygons withshape descriptors using support vector machine(SVM).Finally,based on seed polygons,a regionalgrowing method is proposed to connect the arterial-road features according to the Gestalt concept.Anexperiment shows that this method achieves good and effective arterial road extraction results,reflec-ting the constructed skeleton in urban network.Furthermore,a comparison analysis of high-levelroads attributes was also carried out to confirm the consistency between constructed road levels andour detected arterial roads.