The paper presents a framework for understanding AR learning from three perspectives: physical, cognitive, and contextual. On the physical dimension, we argue that physical manipulation affords natural interactions, thus encouraging the creation of embodied representations for educational concepts. On the cognitive dimension, we discuss how spatiotemporal alignment of information through AR experiences can aid student's symbolic understanding by scaffolding the progression of learning, resulting in improved understanding of abstract concepts. Finally, on the contextual dimension, we argue that AR creates possibilities for collaborative learning around virtual content and in non-traditional environments, ultimately facilitating personally meaningful experiences.