Student access to education levels, tracks or majors is usually determined by their previous performance, measured either by internal exams, designed and graded by teachers in school, or external exams, designed and graded by central authorities. We say teachers grade on a curve whenever having better peers harms the evaluation obtained by a given student. We use rich administrative records from public schools in Catalonia to provide evidence that teachers indeed grade on a curve, leading to negative peer effects. We find suggestive evidence that school choice is impacted only the year when internal grades matter for future prospects.