Three studies were performed to examine the effects of formats of ‘pre-fabricated’ and learner-generated representations on learning outcomes of pupils learning combinatorics and probability theory. In Study I, the effects of different formats on learning outcomes were examined. Learners in five experimental conditions were provided with a simulation-based learning environment. The experimental manipulation concerned the format of the simulations. These were diagrammatical, arithmetical, textual, a combination of textual+arithmetical or diagrammatical+arithmetical. The main finding of the study is that learning from the textual+arithmetical format was most beneficial, in particular with regard to procedural knowledge. Diagrams were found to negatively affect learning and to increase cognitive load. In Studies II and III learners had to work through the same kind of learning environment as in Study I, but they also had to construct representations themselves. The experimental manipulation concerned the format of the representation: graphical (concept map), arithmetical, or textual. In Study II the participants worked alone. It was found that the concept-map and the textual format afforded the creation of domain representations more than the arithmetical format. It was also found that learners engaging in the construction of representations showed higher learning outcomes. This could not be attributed to prior knowledge. Study III was identical except that participants worked in dyads. Again, the concept map and the textual format afforded the creation of representations more than the arithmetical format, although here no differences with regard to learning outcomes were observed between dyads engaging in representation construction and dyads that did not.