In this paper we investigate the existence and the size of territorial differences in Italian students’ mathematical competencies. Our analysis benefits from a new data set that merges the 2003 wave of the OECD Programme for International Student Assessment (PISA) with territorial data collected from several statistical sources and with administrative school data collected by the Italian Ministry of Education. We consider three different groups of educational inputs: individual characteristics (mainly family background), school types and available resources, and territorial features related to labour market, cultural resources and aspirations. In addition to the standard gradient represented by parental education and occupation, we find that student sorting across school types also plays a significant role. Among the local factors measured at province level, we find a significant impact of buildings maintenance and employment probabilities. When accounting for territorial differences, we find that most of the North-South divide (75%) is accounted for by differences in endowments, while the local school production functions account for the remaining fraction.