This paper implements a methodology for estimating poverty in Ecuador, Madagascar and South Africa, at levels of disaggregation that to date have not been available. The methodology is based on a statistical procedure to combine household survey data with population census data, imputing into the latter a measure of per capita consumption from the former. The countries are very unlike each other—with different geographies, stages of development, quality and types of data, and so on. Yet the paper demonstrates that in all three countries the poverty estimates produced from census data are both plausible (in that they match well stratum-level estimates calculated directly from the household surveys) and satisfactorily precise (at a level of disaggregation far below that allowed by household surveys).