The drawing sequence of image contours was optimized in order to improve the robot drawing efficiency. According to the number of discrete points or the size of circumscribed rectangle, contours were divided into two parts: bigger ones and smaller ones. For bigger ones, their drawing priority was proportional with their sizes; for smaller ones, their drawing sequence was transformed into traveling salesman problem. With contour centers as cities to be visited, the shortest path to traverse contour centers was obtained using ant colony algorithm. On this basis, pen contact positions were determined with the nearest neighbor method from the opposite direction of drawing sequence. The experimental results showed that the jump time among contours was shortened substantially after using ant colony algorithm, with the shortest path being reduced to 1/5～1/2 of the original one. The algorithm was applied to robot offline drawing program, resulting in a good balance between computation time and results, which also reduced the workload of complex contour teaching.