Currently, there is no gold standard to estimate national HIV prevalence, and solid data are still lacking. The central focus of this research was to explore factors that may account for differences between HIV estimates obtained from antenatal clinic (ANC) surveillance and those derived from population-based surveys. Using the 2003 Ghana Demographic Health Survey (2003 GDHS), two potential sources of bias were explored. For the ANC surveillance, the lack of representativeness of ANC users, and for the survey, the impact of nonresponse due to HIV test refusal were investigated. Multivariate regression analyses indicated that factors associated with lack of public ANC attendance varied by area of residence. In rural areas, lack of public ANC attendance was associated with older age, having never been married, and region of residence while in rural areas, being a Protestant or having no religion, no occupation, and region of residence were predictors of lack of ANC use. Furthermore, predictors of test refusal varied by gender. Among women, test refusal was positively associated with older age Ga-Adangbe ethnicity, condom use, stigma and region of residence and was negatively associated with parity. Among men, on the other hand, test refusal was positively associated with being formerly married, wealth, knowing prior HIV test results, and region of residence and was negatively associated with education and having an STD. Comparisons of the predictors of HIV and test refusal suggested that overall, test refusal may underestimate the population prevalence. Specifically, refusal was associated with a higher risk profile for women and lower risk profile for men. Among women, older age and Ga-Adangbe ethnicity were associated with both HIV status and test refusal. Among men, higher education was positively associated with HIV status and was negatively associated with test refusal. This research suggests that both ANC surveillance and surveys are subject to bias. ANC surveillance may underestimate the HIV prevalence of pregnant women whereas test refusal may lead to the underestimation of the population prevalence due to the selective refusal of individuals with higher risk to be tested.