Tse and Cavanagh (2000, Cognition 74 B27-B32) showed that Chinese and Americans see opposite illusory line motion in the last horizontal stroke of " " in a stroke-by-stroke presentation. They supposed that the direction of motion perceived by Americans was due to transformational apparent motion; this, in turn, makes the fact that Chinese perceive the opposite direction hard to explain. In this study, prediction of the attention gradient model (Hikosaka, Myauchi, & Shimojo, 1993, Vision Research 33 1219-1240) was tested by varying stimulus onset asynchrony (SOA) between the two critical strokes, so that the predicted direction should be opposite in short (90 ms) and long (900 ms) SOAs, even for Chinese. In addition, two possible top-down factors, script (chirographic, square) and stroke sequence (writing, random), were manipulated. Data from thirty-two Chinese subjects tested individually and 163 tested in groups were collected. The results showed that in the 8 conditions (SOA × font × sequence), only when chirographic characters were presented in writing sequence under the 900 ms SOA condition (exactly that used in Tse & Cavanagh, 2000) did Chinese see the opposite direction of movement from what Americans saw. They see the same direction of movement as Americans in the other 7 conditions. Our data support that attention should be involved in this motion perception, and the top-down factors that make Chinese see the opposite direction were the script and stroke sequence derived from over-learned writing experience.