Cognitive mapping is assumed to be through hippocampus-dependent place learning rather than striatum-dependent response learning. However, we proposed that either type of spatial learning, as long as it involves encoding metric relations between locations and reference points, could lead to a cognitive map. Furthermore, the fewer reference points to specify individual locations, the more accurate a cognitive map of these locations will be. We demonstrated that participants have more accurate representations of vectors between 2 locations and of configurations among 3 locations when locations are individually encoded in terms of a single landmark than when locations are encoded in terms of a boundary. Previous findings have shown that learning locations relative to a boundary involve stronger place learning and higher hippocampal activation whereas learning relative to a single landmark involves stronger response learning and higher striatal activation. Recognizing this, we have provided evidence challenging the cognitive map theory but favoring our proposal.