Previous work on shape extraction and description by slice-to- slice contour matching will be extended to multiple objects with branching structures. Both surface fitting and volumetric modeling will be explored, with emphasis on high-level operations and recognition of "obvious" correspondences. Both spatial (or temporal) position and 2-D shape understanding will thus be used to attack the 3-D correspondence problem in a biological domain (retinal microscopy) where shapes are complex and neither size nor orientation is constant from one slice to the next.