In 2008 UNAIDS global report on AIDS, the number of people living with HIV in sub-Saharan Africa was estimated at 22 millions in 2007 and 20.4 millions in 2001, while in the 2002 report, the same estimation for 2001 was 28.5 millions. Changes in UNAIDS reflects evolutions of data sources and methods used for the estimates. Sentinel surveillance of pregnant women attending antenatal clinics (ANC) was developed in 1980's with the WHO recommendation of unlinked anonymous testing approach. The objective was not to be representative but to monitor trends. In the 1990's, as ANC data were available, they were used by EpiModel, a model developed by WHO for HIV prevalence estimates from 1992 to 2000. In 2002, a new epidemiological model called EPP (Estimation and Projection Package) was developed by the UNAIDS Reference Group on Estimates, Modelling and Projections, which, in countries with a generalised epidemic, is still based on ANC data collected over time. Since 2001, many countries have conducted national population-based surveys (NPS). Their results have often diverged from estimates based on ANC data. This was explained by the under-representation of rural clinics in sentinel surveillance and relative small participation rate in NPS. Since, several studies have shown that the impact of several biases (participation rate, non-household members, serological window of tests...) in NPS remains relatively low. NPS constitute a good indicator of HIV prevalence level. If pregnant women can be locally representative of the general population, at the national scale, it depends of the localization of selected clinics. But ANC provides data over time, which is not the case for NPS. The current approach of UNAIDS consists in estimating HIV prevalence trends from ANC and the level of the epidemics from NPS. But the hypothesis that ANC data are representative of trends still needs to be verified when several NPS will be available for a same country.