Spatial ability tests like mental rotation and paper-folding provide strong predictions of an individual’s achievement in science and engineering. What cognitive skills are involved in them? We use a computational model to analyze these tasks, asking how much information must be processed to perform them. The models demonstrate that in some cases stimuli can be vastly simplified, resulting in consistent performance regardless of stimulus complexity. The ability to produce a scaled-down representation of a complex stimulus may be a key skill underlying high spatial ability.