Pré-publication, Document de travail: This paper tests whether we observe sex-discrimination in teachers' grades, and whether such biases affect pupils' achievement during the school year. I use a unique dataset containing standardized tests, teachers' attributed grades, and pupil's behavior, all three at different periods in time. Based on double-differences, the identification of the gender bias in teachers' grades suggests that (i) girls benefit from a substantive positive discrimination in math but not in French, (ii) girls' better behavior than boys, and their initial lower achievement in math do not explain much of this gender bias. Then, I use the heterogeneity in teachers' discriminatory behavior to show that classes in which teachers present a high degree of discrimination in favor of girls at the beginning of the year are also classes in which girls tend to progress more over the school year compared to boys.