Abstract: Fruit size variations could be used to describe grape growth and detect water deficit stress. To realize non-contact and accurate measurements of overlapping grape fruit diameter, authors proposed a computer vision measurement method based on mathematical morphology. In this method，the false minimum points in the original gradient image were eliminated by detected markers. Watershed transform was applied on the modified gradient image to find accurate contour of the target fruit. The equivalent diameter was calculated by fruit pixels. Experimental results and field applications showed that the segmentation algorithm of the system had higher location accuracy and could provide precise contours of the grape. The measurement system has the advantages of non-contact and high accuracy, and the accuracy of repeatability achieves ±9 μm. The system provides a reference for monitoring grape growth and diagnosing grape water deficit, which is used for grape water-saving irrigation.