Multimedia learning environments commonly comprise different sources of information such as various linguistic as well as static and dynamic pictorial representations. In this context, it is frequently assumed that the (simultaneous) presentation of different and possibly complementary sources of information improves learning. During the last 10 years, however, psychological and educational research have collected extensive evidence that the presentation of different sources of information in multimedia learning environments might not improve and even impede learning. If multimedia learning is to be successful, different sources of information need to be integrated into coherent mental representations. In many cases, however, the presentation and use of different sources of information does not result in the construction of appropriate mental representations. Starting from various potentials and problems of multimedia learning, the theory of structure mapping is proposed as a framework that can be used to design multimedia learning environments that systematically encourage and support the mental integration of multiple sources of information. A multimedia learning environment for statistics is described which has been implemented according to the design principles proposed.