Lack of primary schooling among rural children in developing countries is often attributed to credit constraints and household demand for child labour, implying that direct and indirect costs of schooling are high. Surprisingly few studies have considered the importance of parents' expected returns of investing in their childrens human capital, despite the fact that most parents rely on their children for old-age support and subsistence. In this paper, I propose an alternative model for human capital investment based on the household, rather than the individual child, incorporating the fact that parents bear the costs of educating all their children and face uncertainty about the level and share of future returns. This uncertainty can make it optimal for parents to ensure a certain degree of human capital diversification within the household. The model implications allow me to test whether it is the need for diversification or the costs of schooling that dominate the human capital investment decision in rural households. Using extraordinary long panel data from a rural region in Northwestern Tanzania, I find strong empirical evidence of diversification effects for rural sons, but not for rural daughters. Exactly in line with what should be expected for a patrilineal society. This can potentially have far reaching policy implications.