AbstractThere have been suggestions that human navigation may depend on representations that have no metric, Euclidean interpretation but that hypothesis remains contentious. An alternative is that observers build a consistent 3D representation of space. Using immersive virtual reality, we measured the ability of observers to point to targets in mazes that had zero, one or three ‘wormholes’ – regions where the maze changed in configuration (invisibly). In one model, we allowed the configuration of the maze to vary to best explain the pointing data; in a second model we also allowed the local reference frame to be rotated through 90, 180 or 270 degrees. The latter model outperformed the former in the wormhole conditions, inconsistent with a Euclidean cognitive map.