We describe the structure of consumption of households in Mexico where large cash transfers have been made to women in households randomly selected from a poor population. We focus on the effect of the program, identified through the experimental variation, on total expenditure; on total food expenditure (and hence on the share of food); and finally on the composition of the food basket consumed in the household. We find effects that differ from what would be predicted by a unitary model, both in terms of total food share and on the composition of the diet. We also find that we cannot reject a simple collective model where decision power depends on the share of resources.