This paper is the first to provide, in the European context, credible causal estimates for the impact on educational achievement of a means-tested programme that subsidises school lunches. We use administrative data from the city of Barcelona for the whole universe of applications to the programme. Using a Regression Discontinuity Design, we measure the effect of receiving a partial subsidy, as opposed to none; meanwhile a Difference-in-Differences (DiD) strategy allows us to account for the effect of receiving a full subsidy, compared to a partial one. Our results indicate an overall positive effect of the subsidies on educational achievement across all the subjects analysed, with statistically significant estimates only for Catalan language. Heterogeneous results show that those who benefit most are boys who attend large schools and have peers who are, on average, more socio-economically advantaged. The opportunity for subsidy recipients to spend more time at school during lunch, and to communicate and socialise with wealthier and linguistically more competent children is the mechanism that lies behind our main findings.