External representations (ERs), such as diagrams, equations, graphs, etc. are central to the practice and learning of science, mathematics, and engineering, as the phenomena and entities studied in these domains are often not available for direct perception and action. The ability to generate and use ERs in a domain in an integrated fashion, as well as perform transformations on the ERs, is termed representational competence (RC). Many learning difficulties are attributed to difficulties in achieving RC, particularly integration of ERs. RC thus presents a fundamental cognitive difficulty that cuts across different disciplinary domains, making it critical to develop teaching-learning strategies that help learners develop RC. Most accounts of RC are grounded in the classical information processing model of cognition. In this model, a learner experiences high cognitive load during ER integration, as she tries to ‘extract’ information from ERs, internalize this information in the mind, and translate or process it to establish connections between the ERs. This characterization reduces the content of ERs to information, and treats ERs as ‘vehicles’ of information. This approach therefore does not seek to provide detailed accounts of the role played by ERs in cognition, and does not examine the cognitive mechanisms supporting integration of different ERs. Models based on this framework thus focus on processing cognitive load, and do not provide specific instructional design principles for effective development of RC. Recent theories of cognition have moved away from this type of information processing models, to develop 'field' theories such as distributed and embodied cognition. These accounts suggest that ERs, and a learner’s interaction with them, play a constitutive role in her learning of concepts. I extend this approach in this dissertation, to develop a theoretical model of the cognitive mechanisms underlying ER integration. This model focuses on how the cognitive system interacts with external representations, and the way integration abilities develop through this interaction. This mechanism model predicts that (i) the development of the ER integration ability would result in a reorganization of the sensorimotor system, and (ii) sensorimotor interaction would support ER integration and its development. To test these predictions, I developed two empirical studies, one based on ER categorization tasks and eye tracking, and the other based on the design, development, and testing of an enactive new media intervention. The results from these studies broadly support the theoretical model. Based on these results, I outline some of the broader implications of the model and possible learning interventions.