This paper, based on a large-scale field experiment, tests whether a one-hour exposure to external female role models with a background in science affects students' perceptions and choice of field of study. Using a random assignment of classroom interventions carried out by 56 female scientists among 20,000 high school students in the Paris Region, we provide the first evidence of the positive impact of external female role models on student enrollment in STEM fields. We show that the interventions increased the share of Grade 12 girls enrolling in selective (male-dominated) STEM programs in higher education, from 11 to 14.5 percent. These effects are driven by high-achieving girls in mathematics. We find limited effects on boys' educational choices in Grade 12, and no effect for students in Grade 10. Evidence from survey data shows that the program raised students' interest in science-related careers and slightly improved their math self-concept. It sharply reduced the prevalence of stereotypes associated with jobs in science and gender differences in abilities, but it made the underrepresentation of women in science more salient. Using machine learning methods, we leverage the diversity of role model profiles to document substantial heterogeneity in the effectiveness of role models and shed light on the channels through which they can influence female students' choice of study. Results suggest that emphasis on the gender theme is less important to the effectiveness of this type of intervention than the ability of role models to convey a positive and more inclusive image of STEM careers.