Gender stereotypes in the natural sciences may discourage girls from pursuing STEM fields, thus contributing to the differential STEM pathways of males and females. This paper exploits quasi-experimental data from a vignette study to investigate teachers' gender bias in STEM at the transition to upper secondary school in Denmark--a key stage in students' educational trajectories. I investigate if teachers have a higher probability of recommending a STEM track to a (vignette) male student compared with a (vignette) female student and if teachers' STEM recommendations interact with their demographic characteristics. Results show that, while there is a gender gap of 10 percentage points in the likelihood of being recommended a STEM track, the difference is not statistically significant. Furthermore, teachers' gender bias is influenced by the teacher's own gender and cultural capital. Consequently, the paper shows that teachers' gender bias varies with teachers' demographic characteristics and teachers with high levels of cultural capital can push back against gender stereotypes in STEM.