<p>The current study presents a theory-driven, person-centered analysis of the relationship between different modes of cognitive and metacognitive engagement and immediate and delayed knowledge in chemistry. In a classroom study with high ecological validity, we inferred cognitive and metacognitive engagement from log files of secondary school students’ (<i>N</i> = 235) activities completed within the context of a digital chemistry textbook. We used the ICAP framework to classify cognitive engagement as either passive, active, or constructive, and we extend the framework to include metacognitive engagement. A latent profile analysis identified six distinct latent profiles, characterized by different distributions of student cognitive and metacognitive engagement while learning with the digital textbook: Disengaged; Low Metacognition; Disfavors Active; Moderate Engagement; Metacognitive, but not Constructive; and High Overall Engagement. We found that these profiles were predictive of immediate (one week later) and delayed (six to eight weeks later) chemistry knowledge, even after controlling for prior knowledge. Post-hoc tests indicated that “High Overall Engagement” and “Metacognitive, but not Constructive” were the best performing profiles in both immediate and delayed learning outcomes. Interestingly, profile membership was not predicted by self-reported chemistry self-concept or interest, measured prior to learning with the digital textbook. Our findings provide a novel extension of the ICAP framework and highlight the role of metacognitive engagement for lasting learning in digital learning environments.</p>