Publisher Summary This chapter introduces a new geometric-stochastic expert system approach to building models for and estimating main roads in aerial images. The emphasis of the chapter is on the automatic extraction of main roads when road curvature, width, image intensity, and edge strength can vary considerably, and when a barrier along the road center may or may not be present. The approach is general, and it can be extended to handle the full range of road image variability. The approach is to build geometric-stochastic models for representing road images, and then to use maximum a posteriori probability estimation for estimating the road boundaries and other important features in an image. The modeling approach forces the designer to model all significant phenomena—and the model is generative—so that its representational power can be assessed. The map estimation provides for the most accurate road finding. Global map estimation is realized in a computationally reasonable way by using dynamic programming to search small windows to obtain initial road candidates, and then using dynamic programming again with small windows to obtain global estimates. Contextual information can be built into this geometric-stochastic modeling to handle the full range of imagery and artifacts to be expected.