Accurately inferring three-dimensional (3D) structure from only a cross-section through that structure is not possible. However, many observers seem to be unaware of this fact. We present evidence for a 3D amodal completion process that may explain this phenomenon and provide new insights into how the perceptual system processes 3D structures. Across four experiments, observers viewed cross-sections of common objects and reported whether regions visible on the surface extended into the object. If they reported that the region extended, they were asked to indicate the orientation of extension or that the 3D shape was unknowable from the cross-section. Across Experiments 1, 2, and 3, participants frequently inferred 3D forms from surface views, showing a specific prior to report that regions in the cross-section extend straight back into the object, with little variance in orientation. In Experiment 3, we examined whether 3D visual inferences made from cross-sections are similar to other cases of amodal completion by examining how the inferences were influenced by observers' knowledge of the objects. Finally, in Experiment 4, we demonstrate that these systematic visual inferences are unlikely to result from demand characteristics or response biases. We argue that these 3D visual inferences have been largely unrecognized by the perception community, and have implications for models of 3D visual completion and science education.
Keywords: Amodal completion; Perception of 3D volumes; STEM education; Spatial reasoning.