This chapter argues that to understand the ways that multiple representations should be designed to support learning, we need to consider the pedagogical functions that they play alongside their structural form. Multiple representations play a complementary role when learners exploit differences in computational properties or information by switching between representations and selecting the appropriate representation for the task at hand. Constraining benefits can be achieved when we support learners’ understanding of a second representation by relating it to a familiar representation. Learners can gain deeper understanding when they abstract over multiple representations to achieve insight into the nature of the representations and the domain. This chapters reviews studies that have used multiple representations for these purposes and identifies some of the circumstances that influence the effectiveness of using multiple representations in these ways.