An approach is presented to extract roads from aerial city images based on the Dempster-Shafer evidence theory. A road model is a priori constructed. Aerial images are divided into sub-blocks from which regions consisting of 8-connected pixels with similar gray scales are obtained, and regions with relatively big areas are selected as candidate road segments. Then Dempster rule is applied to compute the fusion of basic probability assignment functions (BPAF) defined respectively on the features extracted from the candidate road segments and on the original road model. Finally, the BPAF fusion is utilized to find the conclusive road segments and these road segments are connected and pruned to form the de facto roads. Experimental results demonstrate the ability of the D-S evidence theory based approach to accurately extract roads from aerial city images.