In phase 1, participants memorized two two-dimensional maps consisting of routes with concrete landmark features and two maps consisting of routes with abstract landmark features and then made distance estimates and mental walks for the routes on each map. Participants estimated significantly longer lengths for routes with concrete features versus abstract features, for routes with four versus two features, and for routes with linear sequences of features versus clustered features. In phase 2, a feature recognition task indicated that participants had significantly greater accuracy and faster response times for concrete features versus abstract features and for features that had appeared in linear arrangements versus clustered arrangements. Our results suggest that the number, type, and configuration of landmark features can distort humans’ memories of path lengths, even when the paths are originally viewed on a simple two-dimensional map rather than encountered through embodied experience.