Cognitive theories on learning with multimedia recommend combining different presentation formats (e.g., text and pictures) in teaching. However, recent research showed that pictures lurk people into trusting the accompanying text, instead of critically studying it. To investigate this we asked 36 physics students to solve mathematical problems, which consisted of a text describing the problem and a statement about the problem that had to be confirmed or rejected (control). The multimedia condition additionally received graphs displaying the same information as the text. Results revealed a bias to confirm statements that were accompanied by a graph irrespective of their correctness. Eye tracking showed that students looked less at the text and the problem statement when a graph was present. The more students looked at the statement and the more transitions they made between the statement and the graph, the better they performed. Verbal data showed that students heavily relied on the graphs: when the graph itself was correct, but the statement was not, students judged the statement as correct referring to the graph. Thus, the mere presence of pictures is not sufficient. Instead, they need to be carefully integrated with the problem statement to improve performance.