Sub-national estimates of HIV prevalence can inform the design of policy responses to the HIV epidemic. Such responses also benefit from a better understanding of the correlates of HIV status, including the association between HIV and geographical characteristics of localities. In recent years, several countries in Africa have implemented household surveys (such as Demographic and Health Surveys) that include HIV testing of the adult population, providing estimates of HIV prevalence rates at the sub-national level. These surveys are known to suffer from non-response bias, but are nonetheless thought to represent a marked improvement over alternatives such as sentinel surveys. At present, however, most countries are not in a position to regularly field such household surveys. This paper proposes a new approach to the estimation of HIV prevalence for relatively small geographic areas in settings where national population-based surveys of prevalence are not available. The proposed approach aims to overcome some of the difficulties with prevailing methods of deriving HIV prevalence estimates (at both national and sub-national levels) directly from sentinel surveys. The paper also outlines some of the limitations of the proposed approach.