When vulnerable population groups are numerically small –as is often the case, obtaining representative welfare estimates from non-purposive sample surveys becomes an issue. Building on a method developed by Elbers, Lanjouw and Lanjouw (2003) it is shown how, for census years, estimates of income poverty for small vulnerable populations can be derived by combining sample survey and population census information. The approach is illustrated for Uganda, for which poverty amongst households with disabled heads is determined. This is possibly the first time that, for a developing country, statistically representative information on income poverty amongst disabled people is generated.