To determine whether teachers' reports of student academic performance can suffice for research purposes by comparing it with a curriculum-based standardised test.In this longitudinal cohort study of children born at risk of neonatal hypoglycaemia, teachers' global assessment of student performance was compared with assessment tools for teaching and learning (asTTle) at 9-10 years. Performance on asTTLe was rated as being below, at or above that expected on the national curriculum for year and term of schooling. Teachers similarly rated the child's performance against the national curriculum.Of 125 children assessed, 104 had paired data for analysis. On asTTLe, 28% were rated below, 55% at and 17% above the expected curriculum level in reading and 24, 54 and 22%, respectively, in mathematics. Equivalent teacher ratings were 23, 58 and 19% in reading and 36, 55 and 9% in mathematics, respectively. However, there was limited agreement between asTTle and teacher rating of achievement in reading (κ = 0.23 (95% confidence interval 0.07-0.40)) and no significant agreement in mathematics (κ = 0.07 (95% confidence interval -0.09-0.22)). Only 45% of children performing below the curriculum level in reading and 52% in mathematics were correctly identified by teachers.In cohort studies, teacher ratings cannot substitute standardised educational testing.