We discuss a fully automatic technique for extracting roads in urban environments. The method has its bases in a vegetation mask derived from multispectral IKONOS data and in texture derived from panchromatic IKONOS data. These two techniques together are used to distinguish road pixels. We then move from individual pixels to an object- based representation that allows reasoning on a higher level. Recognition of individual segments and intersections and the relationships among them are used to determine underlying road structure and to then logically hypothesize the existence of additional road network components. We show results on an image of San Diego, California. The object-based processing component may be adapted to utilize other basis techniques as well, and could be used to build a road network in any scene having a straight-line structured topology.