Engagement in digital reading assessments has been underexplored in spite of its importance for assessment and instructional design. In this study, we used different engagement indicators in two ePIRLS tasks (n[subscript 1] = 14,324, n[subscript 2] = 15,026) to identify engagement profiles, and interpreted them using the three-dimensional model of engagement: affective, behavioral, and cognitive. In both tasks, we found profiles representing high, moderate, and low levels of engagement, as well as profiles with high behavioral-cognitive engagement but low affective engagement, high cognitive engagement, and high behavioral but low cognitive engagement. The students who took both tasks (n[subscript 3] = 2658) were likely to be grouped into similar profiles in both of them, demonstrating the profiles' stability and generalizability. These profiles significantly predicted students' performance, self-efficacy, and reading engagement outside of school (all p < 0.001), with especially strong effects on performance ([eta subscript 1 superscript 2] = 0.22, [eta subscript 2 superscript 2] = 0.17). While most groups' performance was as expected, high behavioral-cognitive and low affective engagement students had the lowest reading engagement outside of school. This study reveals the unique reading engagement patterns that readers exhibit in digital contexts, contributing to the literature on reading engagement. From the perspective of large-scale assessments, the study demonstrates the complex relationship between different forms of engagement and performance, suggesting that response time alone might not be enough to identify disengaged participants.