The debate over the impact of education on economic growth has recently led to disagreement when, at the empirical level, the effect of average human capital on economic growth has been found to be weak. With this paper we revisit these results by arguing how different educational attainment levels (rather than the average human capital stock) impact heterogenously different regions’ economic performance. We build and test a catch-up model where technology adoption takes place as a function of each region’s human capital composition. We show for 50 NUTS3 Spanish provinces in between 1965 and 1997, how convergence to the frontier is driven by higher education and, to a lesser extent, by vocational training. Both theoretical and empirical results are alternative to the well known formalization proposed by Vandenbussche, Aghion and Meghir (2006). Severe endogeneity issues, as well as small sample biases, are tackled by using system GMM estimators and the correction proposed by Windmeijer (2005).