Recent research has shown that females make classrooms more conducive to effective learning. We identify the effect of a higher share of female classmates on students' disruptive behavior, engagement, test scores, and major choices in disadvantaged and non-disadvantaged schools. We exploit the random assignment of students to classrooms in early high school in Greece. We combine rich administrative data with hand-collected student-level data from a representative sample of schools that feature two novel contributions. Unlike other gender peer effects studies, a) we use a rich sample of schools and students that contains a large and diverse set of school qualities, and household incomes, and b) we measure disruption and engagement using misconduct-related (unexcused) teacher-reported and parent-approved (excused) student class absences instead of self-reported measures. We find four main results. First, a higher share of female classmates improves students' current and subsequent test scores in STEM subjects and increases STEM college participation, especially for girls. Second, a higher share of female classmates is associated with reduced disruptive behavior for boys and improved engagement for girls, which indicates an increase in overall classroom learning productivity. Third, disadvantaged students - those who attend low-quality schools or reside in low-income neighborhoods - drive the baseline results; they experience the highest improvements in their classroom learning productivity and their STEM outcomes from a higher share of female classmates. Fourth, disadvantaged females randomly assigned to more female classmates in early high school choose college degrees linked to more lucrative or prestigious occupations 2 years later. Our results suggest that classroom interventions that reduce disruption and improve engagement are more effective in disadvantaged or underserved environments.