BACKGROUND: Risk factors for HIV infection can act at one of several causal levels, making interpretation of results problematic. One suggested solution has been a proximate determinants framework, in which risk factors are grouped into "underlying", "proximate" and "biological" determinants. METHODS: A baseline, cross-sectional survey of HIV serostatus and potential risk factors was carried out among 9480 adults in Zimbabwe. Associations were assessed separately for men and women using logistic regression models; data were only included for those who reported sexual debut. The predictive ability of proximate determinants describing both individual and partnership characteristics was assessed along with that of the underlying determinants. The significance of the underlying determinants once adjusted for proximate determinants was then evaluated. Finally, the relationship between the underlying determinants and some of the key proximate determinants was explored. RESULTS: The two most important proximate determinants for men and women were lifetime number of sexual partners and symptoms of sexually transmitted infections (p<0.001). After adjustment for all proximate determinants, some underlying determinants were still significant, particularly age group, marital status and community (p<0.001). CONCLUSIONS: Although proximate determinants could explain the action of many underlying determinants, several of the latter remained significant after adjustment for the proximate determinants. Hence, the proximate determinants were probably not measured completely. An important determinant of an individual's risk of HIV infection is the HIV status of their sexual partners. This was not measured in this survey, and may be related to the individual's age (as a predictor for the age of the partner), marital status and community prevalence. However, it will be measured in a subsequent survey of this cohort.