<p>This systematic review examines model-based reasoning (MBR) in STEM education, focusing on how it is defined, how it works in practice, and how it is measured and taught across classrooms and laboratories. Guided by PRISMA 2020, this review synthesized 146 peer-reviewed studies published between 1980 and 2025 to address four research questions. We report a qualitative synthesis and descriptive frequencies from the findings. First, the literature converges on MBR as an iterative, distributed, and representation-mediated practice that links mental models with external inscriptions, including diagrams, equations, prototypes, code, and simulations, while integrating abductive, inductive, deductive, causal-mechanistic, and computational forms of reasoning. Second, the reviewed studies suggest recurring stage-based patterns in modeling and simulation activities: abductive and analogical reasoning are especially visible during early problem analysis and formulation; deductive, quantitative, and algorithmic reasoning support model construction and execution; diagnostic, inductive, and probabilistic reasoning support verification, validation, and debugging activities. Third, the field broadly agrees on the centrality of iteration, external representations, and collaboration in model-based reasoning activities, while debates persist over primary theoretical emphasis, particularly whether model-based reasoning is best grounded in mental models or distributed cognition; whether reasoning modes should be treated as analytically separable or as hybrid in use; and the extent to which domain-specific standards should guide model evaluation and acceptance decisions. Fourth, the ways MBR is characterized and measured shape what can be claimed about it: micro-level approaches, such as think-aloud protocols and time-stamped coding, capture moment-to-moment strategy use; meso-level approaches, such as computational notebooks, simulation logs, and rubric-based assessments, reveal workflow and representational competence; and macro-level approaches, such as model-evidence link diagrams and portfolios, capture longer-term development over weeks or semesters. These findings position MBR as a useful integrative lens for scientific sensemaking that is applicable across STEM disciplines, while also showing that its meaning and assessment remain shaped by disciplinary, instructional, and methodological context.</p>