In order to improve the recognition accuracy, a method of overlapping tomatoes recognition based on curvature analysis of edge points was presented. Edges of tomatoes were extracted from binary image at first. Then curvature values could be acquired after edge points were sorted counterclockwise. The edge points with abnormal curvature values were abandoned. After circle regression for remaining edges, overlapping tomatoes could be recognized. In order to decrease the negative impact on recognition aroused from environmental lighting variation and occlusion caused by leaves and branches, threshold segmentation method based on normalization aberration, 6 rules for edge recognition and 3 rules for circle regression were adopted. The experimental results of 119 images showed that the recognition accuracy for overlapping tomatoes with slight occlusion was 90.9%, it was 76.9% when the occlusion rate was between 25% and 50%, and it was 23% when the occlusion rate was larger than 50%.