Background: Previous research indicates that students lack sufficient online credibility evaluation skills. However, the results are fragmented and difficult to compare as they are based on different types of measures and indicators. Consequently, there is no clear understanding of the structure of credibility evaluation. Objectives: The present study sought to establish the structure of credibility evaluation of online texts among 265 sixth graders. Methods: Students' credibility evaluation skills were measured with a task in which they read four online texts, two more credible (a popular science text and a newspaper article) and two less credible (a layperson's blog text and a commercial text). Students read one text at a time and evaluated the author's expertise, the author's benevolence and the quality of the evidence before ranking the texts according to credibility. Four competing measurement models of students' credibility evaluations were assessed. Results: The model termed the Genre-based Confirming-Questioning Model reflected the structure of credibility evaluation best. The results suggest that credibility evaluation reflects the source texts and requires two latent skills: confirming the more credible texts and questioning the less credible texts. These latent skills of credibility evaluation were positively associated with students' abilities to rank the texts according to credibility. Implications: The study revealed that the structure of credibility evaluation might be more complex than previously conceptualized. Consequently, students would benefit from activities that ask them to carefully analyse different credibility aspects of more and less credible texts, as well as the connections between these aspects.