Computer-based learning environments commonly comprise symbolic as well as static and dynamic pictorial representations, frequently combined with the possibility of modifying them interactively. While multiple, dynamic and interactive external representations have the potential to improve learning in various ways, they also place specific demands on learners, such as the need to process and relate different representations, to control and evaluate interactions with these representations, and to construct coherent mental representations. Because learners frequently are not able to meet these demands, the presentation of multiple, dynamic and interactive representations might not only not improve but might even impede learning. Starting from cognitive load theory as well as from structure mapping theory, we developed support measures that encouraged learners to actively integrate symbolic and pictorial representations and to interact with dynamic pictorial representations in a structured and reflective way. In two experimental studies, the learning effects of the encouraged activities were evaluated. Analyses of variance revealed (1) that the active integration of different representations improved learning significantly and (2) that the structured interaction with different representations specifically increased verbal understanding.