Abstract: An automated method to extracting buildings from the large-scale aerial images is proposed. Based on the image features and object domain knowledges of the buildings,the five key algorithms are developed:orienation based weak edges detection,straight lines extraction by the overcrossing bridge manner,corner analysis and detection,contour reasoning and.verification,contour modification and representation.In this paper,the strategy of "focus of attention", which is often used in the field of Al,is utilized to control and organize the extracting process. Considering the efficiency and flexibility of the method,the symbol operating is integreted with the image operating.The remarkable processed results are lastly obtained that the method can be capable of extracting the flat or slope roofs type of the buildings interfered with the surroundings noisy such as the Shades of trees.