Topographic maps represent a three-dimensional (3D) surface using a system of symbols in two-dimensions (2D). To facilitate students’ understanding of topographic maps, the Augmented Reality (AR) Sandbox reads the elevation of actual sand and projects the topographic map lines onto the surface of the sand. Although over 600 institutions have built AR Sandboxes to help people interpret topographic maps, classroom studies using the AR Sandbox have not found significant gains on topographic map assessments. The present study is a 2 by 2 design testing the affordances of the AR Sandbox in a laboratory setting. In the first level of the study, participants interacted with the AR sandbox (3D) or a regular computer monitor (2D) to give them feedback on five landforms they constructed in the sand. Participants initially constructed the landforms in the sand with the feedback off (i.e., sand without the overlaid projection or monitor). The feedback was then turned on and participants compared and contrasted their landform to the target topographic map. Participants were then asked to modify their landform with the feedback on (continuous), or the feedback was turned off (discrete) during modification. A mixed-ANOVA revealed significant gains on a modified version of the Topographic Map Assessment (TMA-B) from pre- to post-intervention (F(1, 74) = 80.34, p < .001). A significant interaction revealed that participants in the 2D condition had greater gains (M = 2.91, SD = 2.48) than those in the 3D (M = 1.64, SD = 2.07) condition (F(1, 74) = 6.38, p = .014), although both conditions had significant pre- to post-intervention improvement (2D: t(37) = 7.24, p < .001, d = 1.02; 3D: t(39) = 5.01, p < .001, d = 0.64). On average, the discrete feedback groups spent significantly less intervention time (M = 48.3, SD = 16.9) compared to the continuous groups (M = 58.2, SD = 18.1) (F(1, 76) = 6.20, p = .015). The findings suggest that the AR Sandbox does improve topographic map skill for individual students using our approach, and that the most efficient technique engages students in discrete episodes of feedback using the 2D computer monitor.