This chapter describes the design and testing of a biologically inspired vision-based model of spatial memory. Three theories of biological spatial memory are discussed. The Topological Network-map theory is translated into general principles, and two forms of connectionist implementation of these principles are discussed. This is followed by a discussion of planning and map-learning experiments performed with a robot. These experiments reveal problems with the implementation of the view-graph principle. The causes of these problems are discussed and solutions proposed.