We investigate possible explanations for the educational gender gap at age 16. We employ a national dataset of matched exam results of the cohort of pupils who took Key Stage 3 tests in 1999 and GCSEs in 2001. Our key result is the sheer consistency of the gender gap, across both the attainment and the ability distribution, with regard to both raw outcomes and value added. It is primarily driven by performance differentials in English. The generality of the gender gap suggests its source is not within-school practice, which means that policy directed at improving such practice may be misplaced.