Virtual models are increasingly employed in STEM education to foster learning about spatial phenomena. However, the role of user interface design and spatial ability in moderating learning and performance with virtual models are not yet well understood. In order to examine the effects of virtual interface design and spatial ability, two experiments were conducted in which participants solved spatial organic chemistry problems using a virtual model system. Two aspects of the virtual model interface were manipulated: display dimensionality (3D stereo vs. mono) and the location of the hand-held device used to manipulate the virtual molecules (colocated with visual cues vs. displaced). The experimental task required participants to interpret and manipulate the spatial structure of organic molecules in order to align various orientations and configurations depicted by diagrams in Experiment 1 and three-dimensional representations in Experiment 2. Individual differences were measured by two spatial ability tests and performance metrics included solution time and accuracy. Colocating the interaction device with the virtual image led to improved performance in both experiments and providing stereo viewing cues led to improved performance in Experiment 2. Individual differences analysis revealed that the effect of colocation on performance was moderated by spatial ability in Experiment 1, and the effect of providing stereo viewing was moderated by spatial ability in Experiment 2. The results were in line with the ability-as-compensator hypothesis: participants with lower ability uniquely benefited from the treatment, while those with higher ability were not affected by stereo or colocation. The findings of this work contribute to the aptitude-treatment-interaction literature and demonstrate a case of increased perceptual fidelity in a virtual model system differentially benefiting individuals with low spatial ability.