In previous research, we found that twelve hours of 3-D spatial training, compared to a randomized control condition, improved the spatial skills and physics exam scores of gifted science, technology, engineering, and mathematics (STEM) undergraduates (n = 55) directly after training. This paper reports on longitudinal findings of this training study. After eight months, training differences did not exist for spatial skills, physics grades, or physics self-efficacy. Large gender differences, favoring males, existed for some spatial skills, physics self-efficacy, and physics grades. Correlational analyses found that mental rotation performance, not spatial working memory, predicted physics self-efficacy and some physics learning outcomes. These results suggest that sustained exposure to spatially enriching activities over several semesters or years may be necessary to address concerning gender gaps in spatial skills among those most likely to pursue advanced educational and occupational positions in physics.