This paper uses rich administrative data from Chile to estimate teacher value added (TVA) on test scores and on an educational attainment index. We allow each teacher to have a different TVA for male and female students and show that differences in TVA explain an important part of the gender gaps we observe in test scores and postsecondary education trajectories. The gaps in gender-specific teaching effectiveness are especially pronounced in mathematics. Indeed, eliminating within-teacher differences in math test score VA would reduce the gender gap in math performance by 67%. We explore what could be behind these gaps in gender-specific TVA and find no significant differences in what makes teachers effective for male and female students. We do find, however, significant associations between teacher characteristics--e.g., gender and performance in the college admission exam--and practices--e.g., paying attention to low-performing students, congratulating students who improve, and having a good relationship with students--with teacher effectiveness. Finally, we also show that math teachers tend to be biased in favor of male students and that teachers with smaller gender biases are more effective for both, male and female students.