This paper assesses the role of social affiliation, measured by caste, in shaping investments in child health. The special setting that we have chosen for the analysis – tea estates in the South Indian High Range – allows us to control nonparametrically for differences in income, access to health services, and patterns of morbidity across low caste and high caste households. In this controlled setting, low caste households spend more on their children's health than high caste households, reversing the pattern we would expect to find elsewhere in India. Moreover, health expenditures do not vary by gender within either caste group, in contrast once again with the male preference documented throughout the country. A simple explanation, based on differences in the returns to human capital across castes in the tea estates is proposed to explain these striking results.