Stereoscopy's potential as a tool for science education has been largely eclipsed by its popularity as an entertainment platform and marketing gimmick. Dozens of empirical papers have been published in the last decade about the impact of stereoscopy on learning. As a result, a corpus of research now points to a coherent message about how, when, and where stereoscopy can be most effective in supporting science education. This position paper synthesizes that research with examples from three studies recently completed and published by the authors of this paper. Results of the synthesis point towards generally limited successful uses of stereoscopic media in science education with a pocket of potentially beneficial applications. Our position is that stereoscopy should be used only where its unique properties can accommodate specific requirements of understanding topics and tasks--namely visualizations where the spatial sense of depth is germane to conveying core ideas and cognitive load is high. Stereoscopy's impact on learning is also related to the spatial ability of the viewer. More research is needed on the effect of novelty, long-term learning and possible learning differences between the various methods of implementing stereoscopy.