Perception and understanding of the orientations of objects depend on structural descriptions of spatial relations. As a consequence, altering the orientation of a perceived object can profoundly affect reasoning with respect to the object's shape or rotation. To learn more about structural descriptions in spatial understanding, we are investigating the processes by which people learn to correctly predict the outcomes of rotations that initially are very difficult. The method incorporates a virtual reality (VR) computer system with stereoviewing, photorealistic graphics, smooth motion, and user interaction with the scene. The object is a transmitter dish attached to a shaft; the dish is at different orientations to the shaft and the shaft is at different orientations to the environment. Each trial in the experiment includes three self-paced phases designed to allow optimal learning. First, the participant reasons about the direction the dish will face after a rotation of the shaft and indicates this direction by adjusting the direction of an arrow. Second, the participant receives written feedback and a demonstration of the correct answer in VR. Third, the participant rotatates the assembly and observes the motion. One version of this task provided insightful computer visualization that made correct reasoning easy. Interestingly, although this task required attention to the motion and user interaction with the assembly, participants failed to learn the new structural relations. In a different task, the displays did not include the added visualization. This task was initially difficult, but participants were able to learn the structural relations necessary to reason accurately. There were large individual differences in the overall level of performance. Three psychometric variables accounted for 78% of this variance: spatial ability, fluid intelligence, and mastery motivation. Spatial learning in VR transferred to reasoning with real objects.