We use a large and rich administrative household panel data set to estimate labour supply responses for a large number of subgroups in the Netherlands. The identification of the parameters benefits from a major tax reform in the data period. We uncover large differences in behavioural responses. In particular, we find differences in labour supply responses between households with and without children that are much bigger than suggested by previous studies that had to pool these household types in the estimation of preferences. An efficient tax-benefit system should take the substantial heterogeneity in behavioural responses into account.