The reliability of the household consumption-based (Engel curve) methodology in detecting gender bias has been called into question because it has generally failed to confirm bias even where it exists. This article seeks to find explanations for this failure by exploiting a data set that has educational expenditure information at the individual level and also, by aggregation, at the household level. I find that, in the basic education age groups, the discriminatory mechanism in education is via differential enrollment rates for boys and girls. Education expenditure, conditional on enrollment, is equal for boys and girls. The Engel curve method fails for two reasons. First, it models a single equation for the two-stage process. Second, even when we make individual- and household-level expenditure equations as similar as possible, the household-level equation still fails to "pick up" gender bias in about one-third of the cases where the individual-level equation shows significant bias. This article concludes that only individual-based data can accurately capture the full extent of gender bias.