Abstract: In this paper,a new method of quasi-automatic extraction of zonal roads is put forward.Based on research on roads' feature in high resolution RS images,this paper uses two seed points chosen on road surface to automatically search original information of roads;Then,we can establish standard road template,and match it with new road templates got by moving and rotating standard template to compute new direction,center point and edge points of next road segment;Through constantly updating road standard template,direction and center point,automatically obtain a series of edge points distributing on the two-edge of road;Use Snake Model to dynamically adjust edge points and accurately position them on the true edge;Lastly,solve B-spline curve of control points to denote the two borders.The results show the research is effective and viable;Meanwhile,the research makes great advance in the degree of automation,efficiency and accuracy.