An automatic detection method for external features of grafting seedlings based on mathematical modeling was studied. The detecting items included growth status (straight or curved, bending direction), cotyledons parameters (cotyledon flare angle, cotyledons flare spans), hypocotyls parameters (curvature, hypocotyl length, hypocotyl shaft), and other external parameters. First, image preprocessing was used to extract the binary image. Then, a reference point and its position were determined by statistic of horizontal pixels. Next, growth status was decided by a set inclination angle of the minimal bounding rectangle of the hypocotyl and its width. After that, the cotyledons span was calculated by the distance of the two cotyledons endpoint, and the cotyledons angle was computed by the angle between two lines that fitting with the bottom of flat cotyledons. Finally, the stem length and the coarse strains were obtained by doing slope compensations to two sections of stem separately which was divided at the point with maximum curvature. Results were compared with manually measured data, and shown that the coefficients of plant height, plant coarse, and cotyledon span were 0.9351, 0.8999 and 0.9034, respectively. And the relative errors of them were less than 7%, 5% and 7%, while the absolute errors of them were less than 4 mm, 0.2mm and 6mm, respectively.