Is exposing students to role models an effective tool for diversifying science, technology, engineering, and mathematics (STEM)? So far, the evidence for this claim is mixed. Here, we set out to identify systematic sources of variability in STEM role models' effects on student motivation: If we determine "which role models" are effective for "which students," we will be in a better position to maximize role models' impact as a tool for diversifying STEM. A systematic narrative review of the literature (55 articles) investigated the effects of role models on students' STEM motivation as a function of several key features of the role models (their perceived competence, their perceived similarity to students, and the perceived attainability of their success) and the students (their gender, race/ethnicity, age, and identification with STEM). We conclude with four concrete recommendations for ensuring that STEM role models are motivating for students of all backgrounds and demographics--an important step toward diversifying STEM.