The predictive power of polygenic scores for some traits now rivals that of more classical phenotypic measures, leading to suggestions that polygenic scores offer a potentially useful tool for genetically informed policy. However, it is not well understood how much information polygenic scores convey for complex social traits such as education over and above phenotypic data that are available or easily measured. Using data from a UK cohort study we investigate the accuracy with which polygenic scores for education predict pupil’s realised attainment. We test their use as standalone predictors and conditional on phenotypic data that is available to or could be easily and cheaply collected by schools. In our sample, children’s polygenic scores predicted their educational outcomes almost as well as parent’s socioeconomic position or education. There was high overlap between the polygenic score and attainment distributions though, leading to weak predictive accuracy at the individual level. Conditional on prior attainment, polygenic scores were not predictive of later attainment. Our results suggest that while polygenic scores are informative for identifying group level differences in education, they currently have very limited use in predicting how well an individual will perform