I use a combination of blind and non-blind test scores to show that middle school teachers favor girls when they grade. This favoritism, estimated in the form of individual teacher effects, has long-term consequences: as measured by their national evaluations three years later, male students make less progress than their female counterparts. Gender-biased grading accounts for 21 percent of boys falling behind girls in math during middle school. On the other hand, girls who benefit from gender bias in math are more likely to select a science track in high school.