Abstract This paper aims to quantify the effects of different types of sexual risk behaviour on the spread of HIV in South Africa. A mathematical model is developed to simulate changes in numbers of sexual partners, changes in marital status, changes in commercial sex activity, and changes in the frequency of unprotected sex over the life course. This is extended to allow for the transmission of HIV, and the model is fitted to South African HIV prevalence data and sexual behaviour data. Results suggest that concurrent partnerships and other non-spousal partnerships are major drivers of the HIV/AIDS epidemic in South Africa. (ProQuest: ... denotes formulae omitted.) 1. Introduction Understanding the relationship between sexual behaviour and the risk of HIV infection is critical to the development of HIV prevention strategies. Ideally, policymakers should have detailed knowledge of the relative numbers of new HIV infections currently occurring in different sub-populations and risk groups if they are to respond effectively (Bertozzi et al. 2008; Piot et al. 2008). However, measuring numbers of new HIV infections occurring in different risk groups through surveys is challenging, and, in view of the complexities involved, simulation models that are calibrated to available epidemiological data may provide an acceptable alternative method of assessing the relative significance of different risk behaviours (Bertozzi et al. 2008). Mathematical models can also be used to assess the likely effects of different types of behavioural change (Garnett and Anderson 1995; Korenromp et al. 2000; Bracher, Santow, and Watkins 2004; Hallett et al. 2007), and can play an important role in determining whether observed changes in HIV prevalence are attributable to changes in sexual behaviour (Hallett et al. 2006). However, there are several difficulties associated with using mathematical models to assess the contribution of different sexual behaviours to the transmission of HIV. First, there is often a lack of reliable sexual behaviour data for setting the sexual behaviour parameters in these models. In South Africa, for example, hardly any nationally representative sexual behaviour data were collected prior to 2000 (Eaton, Flisher, and Aaro 2003), and as a result, many early models of the HIV/AIDS epidemic in South Africa were forced to rely on fairly arbitrary assumptions about sexual behaviour (Doyle and Millar 1990; Schall 1990; Groeneveld and Padayachee 1992; Johnson and Dorrington 2006). Even when sexual behaviour data are available, these are often affected by social desirability bias and recall bias. Sexual behaviour data have traditionally been captured in face-to-face interviews (FTFIs), but such interviews have been shown to elicit significantly lower reported numbers of sexual partners when compared with recently developed interview formats that are more impersonal and anonymous (Ghanem et al. 2005; Kissinger et al. 1999; Rogers et al. 2005; Mensch, Hewett, and Erulkar 2003; Gregson et al. 2004). Studies have also shown that prompting individuals with additional questions can lead to enhanced recall of past sexual relationships (Brewer et al. 2005). Although a number of mathematical models have been parameterized on the assumption that individuals report their sexual behaviour accurately in FTFIs (Dunkle et al. 2008; Merli et al. 2006; Oster 2005; Leclerc and Garenne 2007), this assumption is questionable in light of the recent evidence regarding social desirability bias. Other models that are calibrated to sexual behaviour data have allowed for bias, implicitly or explicitly, but have not considered the extent of the uncertainty regarding this bias (Van der Ploeg et al. 1998). A further difficulty in using deterministic models to assess the contribution of different risk behaviours to the spread of HIV is that most deterministic models are not sufficiently detailed to provide meaningful insights into sexual behaviour. …