Journal article: The need for education to help every child has become more important for policy in the US and the UK. Remedial programmes are often difficult to evaluate because participation is usually based on pupil characteristics that are largely unobservable to the analyst. We evaluate programmes for children with 'Special Educational Needs' in England. We show that the decentralized design of the policy generates much stronger differences across schools in access to remediation resources for children with moderate learning difficulties than for children with either no difficulties or severe difficulties. However, these differences are not reflected in subsequent educational attainment – suggesting that the programme is ineffective for children with moderate learning difficulties. Also, we use demographic variation within schools to consider the effect of the programme on whole year groups. Our analysis is consistent with no overall effect on account of the combined direct and indirect (spillover) effects.