This paper describes a rather simple but complete object recognition system that was implemented in one semester as a class project in our course in computer vision. The system accepts video images of three-dimensional objects composed of configurations of toy blocks and delivers a one-word classification (e.g., ``arch'') of the scene. The strategy employed is one of identifying individual blocks, establishing spatial relationships between the blocks, and establishing instances of one block supporting another. The constraints of gravity are thus incorporated into the recognition process. It is designed so that any one student can elect to build one of three modules that comprise the system; the system can thus be implemented by three students in one semester. The modules span the complete range of problems encountered in any machine vision system???from image acquisition to high-level model matching. The entire project can be implemented with readily available hardware.