As the accuracy of wave-front detection of Shack-Hartmann sensors is mainly dependent on the accuracy of centroid detection of a focal spot, this paper proposes a dynamic determination method of array spot centroid detection window based on marked watershed segmentation. This method smoothes the spot image captured by a Shack-Hartmann wave-front sensor based on the marked watershed segmentation and obtains its gradient image, firstly. Then the OTSU method is used to mark the target on the gradient image. Finally, the marked gradient image is segmented by the marked watershed method and the detection window of every spot is determined. Because the detecting window can be matched with the actual size of spot, the affection of the noise on centroid detection is effectively suppressed. The experimental results demonstrate that the precision and stability of centroid detection by this method are higher than that of traditional method. The mean values of standard deviation in every detecting window of statistic multiple images are calculated, which is 0.010 9 , 6 times that of the traditional method (0.073 4). It can satisfy the requirement of Shack-Hartmann wave-front sensors for precision and stability of centroid calculation.