Extracting the linear parameters of the runway is the precondition for airborne vision based auto-landing. Firstly it is required to obtain a lot of short lines by gradient direction grouping on the edge map. Secondly according to the angle between two lines and the distance between the ends of two lines, and the gray average on both sides of lines, a rule to quantitatively characterize the affinity between two short lines is advanced, then the short lines on the same line end-to-end by the value of affinity will be linked. Finally the longest lines are left to choose, and priori knowledge, namely the contrast information of shades between the internal and external runway is applied to distinguish the two lines lying on the two edges of the same runway. The method has been tested valid by many actual images.