"Hands-on" activities play an important, but controversial, role in early science education. In this study we attempt to clarify some of the issues surrounding the controversy by calling attention to distinctions between: (a) type of instruction (direct or discovery); (b) type of knowledge to be acquired (domain-general or domain-specific); and (c) type of materials that are used (physical or virtual). We then describe an empirical study that investigates the relative effectiveness of the physical-virtual dimension. In the present study, seventh and eighth grade students assembled and tested mousetrap cars with the goal of designing a car that would go the farthest. Children were assigned to four different conditions, depending on whether they manipulated physical or virtual materials, and whether they had a fixed number of cars they could construct or a fixed amount of time in which to construct them. All four conditions were equally effective in producing significant gains in learners' knowledge about causal factors, in their ability to design optimal cars, and in their confidence in their knowledge. Girls' performance, knowledge, and effort were equal to boys' in all conditions, but girls' confidence remained below boys' throughout. Given the fact that, on several different measures, children were able to learn as well with virtual as with physical materials, the inherent pragmatic advantages of virtual materials in science may make them the preferred instructional medium in many hands-on contexts.