["Journal of Research in Science Teaching, Volume 63, Issue 4-5, Page 352-371, May 2026. ", "\nABSTRACT\n\nModeling involves a set of scientific practices that pre‐service science teachers struggle with. These practices involve constructing, predicting, testing, and revising models. Engaging meaningfully in modeling practices leads modelers to gain new insights about phenomena. Research suggests that this requires domain‐general metamodeling knowledge and domain‐specific content knowledge. However, research has not yet systematically investigated the role of metamodeling and content knowledge in the modeling practices, particularly regarding which practices benefit from either knowledge facet or their interplay. Addressing this gap, we interviewed 17 pre‐service science teachers selected from a larger quantitative study based on their performance in metamodeling knowledge and modeling practice assessments. We showed participants their answers to the modeling practice assessments and asked them to recall their thinking. We analyzed their thinking processes, focusing on metamodeling and content knowledge application. By comparing high‐ and low‐modeling‐practice performers, we aimed to examine the role of these knowledge facets in modeling practices. Our study extends modeling literature with two main findings: First, both metamodeling and content knowledge supported the participants in engaging in modeling practices, but each practice had specific knowledge requirements. For instance, constructing models was driven by content knowledge, while an interplay between metamodeling and content knowledge supported revising models. Second, the metamodeling knowledge that participants demonstrated in the previous metamodeling knowledge assessment was not related to the metaknowledge they applied during modeling practices. This is best evidenced in testing models, for which participants frequently recalled metaknowledge they had not demonstrated in the previous metamodeling knowledge assessment. Our findings suggest that teaching modeling practices separately is probably beneficial, as this approach enables science educators to better scaffold practice‐specific knowledge requirements. For future research, our results question the meaningfulness of metamodeling knowledge assessments, as the knowledge demonstrated in such assessments does not indicate the knowledge learners actually apply when engaging in modeling practices.\n\n"]