People are able to form the correct mental model of puzzles from either visual or haptic experiences alone and are able to predict haptic properties from visual ones. Findings suggest humans segment scenes into objects without explicit boundary cues by using purely statistical information.