Abstract: The recognition of microstructures such as road junctions in road networks is of importance for multi-scale road modeling and pedestrian navigation. Aiming to resolve deficiencies in the current recognition methodsfor geometric shape description and shape matching with complex road junctions, the paper presents a road junction recognition method based on the classification of roads and starting with recognition and reduction. Firstly, junctions are located by the node cluster density detection. Then the characteristic vectors are built by analyzing and quantifying the sizes, shapes, and attributes of roads; treating this issue as a two-class classification problem for differentiating main roads and the auxiliary sections, solved by using a support vector machine. Using Open Street Map data for experimental verification, our results show that this method can effectively recognize road junctions.