Abstract: To create more realistic models for morphology of plant leaves, an approach to reconstructing the 3D model of a leaf blade from images is proposed in this paper. First, feature points in the edge and midrib of the leaf blade were extracted by using a corner detection algorithm, then the 3D coordinates of feature points were calculated by matching these feature points, the matching combines epipolar constraint and Sift feature descriptor. Second, the edge and midrib of the leaf surface were fitted using B-splines curve, then delaunay triangulation algorithm was used to generate a triangulate representation for the leaf blade. Finally, texture mapping was used to the geometry model for improving the appearance. Experimental results show that the model of cucumber leaf shape, reconstructed by the presented approach, is more natural and realistic.