This paper uses linked household, school, and test score data from Madagascar to investigate the relation of household characteristics and school factors to the cognitive skills of children ages 8-10 and 14-16. In contrast to most achievement test studies in developing countries, the study uses representative rather than school-based samples of children and combines detailed information on school and family background. Schooling of mothers matters far more for learning than schooling of fathers, perhaps reflecting differences in parental time spent with children on schoolwork. Even these effects, however, are significantly attenuated when controlling for choice of residence or school. Skills are also affected by aspects of primary schools, including teacher experience and infrastructure. (Contains 11 notes and 7 tables.)