Abstract: This paper is focused on how to extract linear feature from remote sensing image by watershed transform with the help of the information about the linear object. The method extracting the information and incorporating it into the watershed transform is detailed. The method consists of the following steps. ① The marker point of the linear object is detected from the input image; ② These points are utilized to modify the gradient of the input image, which aims to suppress the regional minimum that does not correspond to the linear object; ③ The modified gradient image is segmented by the watershed transform. The segmented regions that correspond to the linear object were merged to represent the object.