This paper is concerned with the issue of the most cost-effective way of improving access to education for poor households in developing countries. We consider two alternatives: (1) extensive expansion of the school system (i.e., bringing education to the poor) and (2) subsidizing investment in education by the poor (i.e., bringing the poor to the education system). To this end, we evaluate the Programa Nacional de Educación, Salud y Alimentación (PROGRESA), a large poverty alleviation program recently introduced in Mexico that subsidizes education. Using double-difference regression estimators on data collected before and after the program for randomly selected control and treatment households, we estimate the relative impacts of the demand- and supply-side program components. Combining these estimates with cost information, we find that the demand-side subsidies are substantially more cost-effective than supply-side expansions.