We examine the relationship between school attendance and HIV incidence among young women in South Africa. Our aim is to distinguish a causal effect from correlation. Towards this end, we apply three methods to population-based longitudinal data for 2005-2012 in KwaZulu-Natal. After establishing a negative association, we first use a method that assesses the influence of omitted variables. We then estimate models with exclusion restrictions to remove endogeneity bias, and finally we estimate models that control for unobserved factors that remain constant over time. All the three methods have strengths and weaknesses, but none of them suggests a causal effect. Thus, interventions that increase school attendance in KwaZulu-Natal would probably not mechanically reduce HIV risk for young women. Although the impact of school attendance could vary depending on context, unobserved variables are likely to be an important reason for the common finding of a negative association between school attendance and HIV incidence in the literature.