We propose using Stochastic Frontier Analysis to estimate pupils' academic underachievement. We model underachievement as the gap between expected achievement and actual achievement, not due to a learning disability. Our data are a panel for 2,228 Belgian pupils observed over 6 years of primary education. We found that the average underachievement gap is 23.5%. That is, the average pupil does not exploit about one fourth of their potential. Gifted pupils appear to underachieve as much as non-gifted pupils. We also found that class size is a determinant of underachievement. The association between class size and underachievement is non-monotonic, with an underachievement minimum at a class size of about 20 pupils.