The collective approach to household behaviour models the household utility function as the weighted average of the utilities of the individual members of the household. These weights, which measure the relative bargaining power of males and females within the household, are generally regarded as fixed and exogenous. The paper extends the collective approach and estimates a model where the weights are endogenously determined. Estimation is conducted using two different data sets from three Indian states. The paper presents evidence on the key determinants of relative power within the household. We find that relative bargaining power has a statistically significant effect on the budget share of an item and that the effects are typically nonlinear and vary significantly across items. We also find evidence of significant gender bias in educational attainment with the bias running in favour of boys in rural Bihar (indicative of active gender differentiation against girls in that parents are unwilling to spend as much on educating girls as they are for boys). The gender bias actually runs the other way in rural Kerala. There is no gender difference in education expenditure in urban samples. This paper departs from the previous literature in being one of the first to find, using household budget data, significant evidence of gender bias in household spending on education.