We provide estimates for the effect of attending a Faith school on educational achievement using a census of primary school pupils in England. We argue that there are no credible instruments for Faith school attendance in this context. Instead, we partially control for selection into religious schooling by tracking pupils over time and comparing attainments of students who exhibit different levels of commitment to religious education through their choice of secondary school and residence. Using this approach, we find only a small advantage from Faith primary schooling, worth about 1 percentile on age-11 test scores. Moreover, this is linked to autonomous admissions and governance arrangements, and not to religious character of the schools. We then go on to show that our estimates vary substantially across pupil subgroups that exhibit different levels of sorting on observable characteristics into Faith schooling, and provide bounds on what the 'Faith school effect' would be in the absence of sorting and selection. Pupils with a high degree of observable-sorting into Faith schools have an age-11 test score advantage of up to 2.7 percentiles. On the other hand, pupils showing a very low degree of sorting on observables have zero or negative gains. It appears that most of the apparent advantage of Faith school education in England can be explained by differences between the pupils who attend these schools and those who do not.