A novel approach to automated road extraction using a hybrid combination of self-organization and model-based extraction techniques is introduced. We use a self-organized mapping technique to first delineate the medial axis topology of road features. This is accomplished through a local clustering of class-binarized spatial information provided from a region segmentation. Since cluster analysis exemplifies a center-of-gravity solution, it is not sensitive to edge definition. Topological structure is subsequently derived through the application of a graph-theoretic approach to link convergent cluster centers. Taking initialization cues from the centerline extraction results of self-organization, a model-based fitting algorithm is then applied to robustly delineate road segment orientations and widths. Preliminary results demonstrate the ability of this approach to automatically extract road centerline position as well as road segment width and orientation in high spatial resolution urban imagery.