When computer vision was applied to fruit detection, the background and sample of an image could be effectively segmented with the aid of boundary tracking technique. According to the appearance of some fruit, such as pear and apple, a modified boundary tracking method was proposed. In this method, a sample in an operation image that could be binary image or gray image was partitioned into 4 regions. To accelerate the boundary tracking, the priority of boundary tracking directions in each region was given by the appearance of the sample. Lastly, tests of boundary tracking using images of apples, pears, mango etc. were done. The results confirmed that the time of processing a 150 by 150 image of a pear was less than 1 s.