Two experiments tested whether 4-year-old children extract and use geometric information in simple maps without task instruction or feedback. Children saw maps depicting an arrangement of three containers and were asked to place an object into a container designated on the map. In Experiment 1, one of the three locations on the map and the array was distinct and therefore served as a landmark; in Experiment 2, only angle, distance and sense information specified the target container. Children in both experiments used information for distance and angle, but not sense, showing signature error patterns found in adults. Children thus show early, spontaneously developing abilities to detect geometric correspondences between three-dimensional layouts and two-dimensional maps, and they use these correspondences to guide navigation. These findings begin to chart the nature and limits of the use of core geometry in a uniquely human, symbolic task.