This article investigates whether different formats of visualizing information result in different mental models constructed in learning from pictures, whether the different mental models lead to different patterns of performance in subsequently presented tasks, and how these visualization effects can be modified by further external representations during task performance. A total of 80 university students learned from an illustrated text different day times and different dates exist simultaneously on the earth. One half of the participants received the text combined with pictures visualizing the earth as a kind of carpet (carpet pictures), whereas the other half received the text combined with pictures visualizing the earth surface as a circle (circle pictures). After learning, the participants received a test including different kinds of tasks. In both visualization groups, one half of the participants solved the tasks with an additional external representation, whereas the other half solved the tasks without an external representation. The findings indicate that the form of visualization affects the structure of mental models. Different structures of mental models result in different patterns of performance, when individuals solve tasks based only on their mental representations acquired during their previous learning. However, these effects decrease, when further external representations are made available to the learners. The findings are discussed within a broader framework of learning with multiple external representations.