We show that the age structure of human capital matters for economic growth. This question had not been tackled empirically until very recently due to the lack of comparable cross-country data on agespecific educational attainment. We present the newly developed IIASA/VID dataset, which provides consistent information about educational attainment by age and sex for 120 countries for the period 1970-2000 at five year intervals. The dataset has been created making use of the information contained in censuses, labor market surveys, and the Demographic and Health Surveys, coupled with a reconstruction exercise based on multistate demographic methods of backward projection. Using the framework of the Benhabib-Spiegel (1994) model, we show that improvements in education in young age groups are particularly important for explaining economic growth in developing countries. Furthermore, we show that the use of education data disaggregated by age improves out-of-sample economic growth predictions significantly.