First-degree course grades for a cohort of social science students are matched to their instructors, and are statistically decomposed into departmental, course, instructor, and student components. Student ability is measured alternatively by university acceptance scores, or by fixed effects estimated using panel data methods. After controlling for student ability, course characteristics, and instructor fixed effects, departmental grade differentials range over 7 points out of 100. Instructors who teach in more than one department grade more generously in departments that award higher grades, suggesting that differential grading policy is set by departments and does not result from unobserved differences in instructor quality and teaching material. Finally, student fixed effects, which measure ability at university, are correlated to 0.41 with their university entrance scores, which measure ability prior to university. This suggests that university entrance scores are poor predictors of student performance in higher education.