This paper is the first to meta-analyse the literature on the causal effects of school-meal programmes on children's behavioural, health and educational outcomes in developed countries, while addressing potential publication bias and heterogeneity between studies. We create a sample of 2,821 estimates from 42 studies and gather 59 aspects reflecting the context in which each estimate was obtained, including type of data, programme characteristics, student population, estimation method and publication quality, among others. We employ both linear and non-linear techniques to correct for publication bias, and we use Bayesian model averaging to study heterogeneous effects and address model uncertainty. The results are consistent with small publication bias - with the exception of studies devoted to analysing test scores, which appear more selective when reporting results. Once publication bias is accounted for, we find that school-meal programmes in high-income economies have minimal impact on students' behaviour, health and education. Our heterogeneity analysis documents the fact that means-tested programmes and breakfast initiatives yield the greatest benefits for children's outcomes.