This article reports on "MaterialSim", an undergraduate-level computational materials science set of constructionist activities which we have developed and tested in classrooms. We investigate: (a) the cognition of students engaging in scientific inquiry through interacting with simulations; (b) the effects of students programming simulations as opposed to only interacting with ready-made simulations; (c) the characteristics, advantages, and trajectories of scientific content knowledge that is articulated in epistemic forms and representational infrastructures unique to computational materials science, and (d) the principles which govern the design of computational agent-based learning environments in general and for materials science in particular. Data sources for the evaluation of these studies include classroom observations, interviews with students, videotaped sessions of model-building, questionnaires, and analysis of artifacts. Results suggest that by becoming "model builders," students develop deeper understanding of core concepts in materials science, and learn how to better identify unifying principles and behaviors within the content matter.