This paper utilises an educational production function approach on post-apartheid data that include both schooling and community-level information, in order to empirically estimate the key determinants of Grade 12 pass rates in 2000. Quantile regression techniques are applied, allowing for more nuanced information. The key results are, firstly, that the pupil–teacher ratio is insignificant in explaining pass rates for schools below the 95th percentile of the school performance distribution. Secondly, the impact of resources on performance is not strong and, where there is a significant effect, it is highly dependent on the resource in question and the metric utilised for the dependent variable. Thirdly, knowledge infrastructure may be important to understand the absolute and relative performance of schools. Fourthly, proxy variables for teacher and parent characteristics are strongly significant, and the former should probably be a priority focus for any policy programme aimed at improving Grade 12 performance levels in South Africa.