A model of the hippocampus as a “cognitive graph” is proposed. It essentially considers the hippocampus as an heteroassociative network that learns temporal sequences of visited places and stores a topological representation of the environment. Using place cells, head-direction cells, and “goal cells”, we propose a biologically plausible way of exploiting such a spatial representation for navigation, which does not require complicated graph search algorithms. Simulations show that the resulting animat is able to navigate in continuous environments that contain obstacles. Furthermore, we make experimental predictions on simultaneous recordings of multiple cells in the rat happocampus.