This paper uses a large dataset combining census, household survey and budgetary data for nearly 4.000 Brazilian municipalities to estimate the impact of government spending on education and health outcomes. We deal with the multi-dimensional nature of the population’s social status by estimating structural equation models with latent variables using a limited-information two-stage least square (2SLS) estimator. Robustness of the baseline regressions to heterogeneity in the data is assessed on the basis of quantile regressions. The main empirical findings are that government spending is a powerful determinant of education outcomes, but this is not the case for health, and that spending on non-education programmes are also at least as important. In addition, there appears to be scope for gains in economies of scale in the provision of education and health care services, at least for selected segments of the conditional distribution of social outcomes. Finally, there are cross-sectoral effects in service delivery: health (education) outcomes affect the population’s education (health) status. This Working Paper relates to the 2009 OECD Economic Survey of Brazil (www.oecd.org/eco/surveys/brazil).