I examine the problem of poverty mapping. I explain the difficulties of small area estimation arising from data sparseness and argue that any attempt to overcome this fundamental problem has to take the form of a conditional homogeneity assumption.