Background: Women in sub-Saharan Africa carry a disproportionate burden and risk of HIV. In this context, women at high HIV risk are not a discrete group, rather on a spectrum of risk caused by a multitude of behavioural, economic, structural, cultural and geographic factors. Pre-exposure prophylaxis (PrEP) is a promising new HIV prevention method, effective at reducing HIV risk when adhered to. However, the results of PrEP trials and implementation studies to date reveal challenges in women’s programme retention and drug adherence. There are also concerns that behavioural disinhibition (reductions in condom use) following the introduction of PrEP may further limit its ability to avert infections. In the context of HIV resource limitations and decreasing donor budgets, this thesis seeks to use mathematical modelling to assess strategies for PrEP scale-up for women across a spectrum of risk in sub-Saharan Africa, accounting for heterogeneities in HIV risk factors and PrEP programme outcomes. Considering the challenges faced by policy makers in using mathematical models to guide decision making, often considered to be complex ‘black boxes’, this thesis also sets out to assess the contexts in which simple models are sufficient to guide policy making around the introduction of a new HIV prevention intervention. Methods: This thesis adopts mathematical modelling approaches to inform HIV policy making. First, a simple static model of HIV risk to female sex workers is developed and used to assess the impact of behavioural disinhibition on PrEP’s ability to avert HIV infections. The static model formulation is then evolved to incorporate dynamic effects to account for the downstream effects of population interactions. The models account for heterogeneities in women’s HIV risk factors and PrEP programme outcomes, and the low levels of PrEP programme retention and adherence reported in studies. The outcomes of the static and dynamic model formulation are compared over different time horizons and epidemic contexts, to contribute to understanding around the importance of modelling complexity to inform HIV policy. Finally, the static model is refined to represent women across a more broadly defined spectrum of risk: women 15-24 years, 25-34 years, 35-49 years and female sex workers. The models are parameterised to case study countries spanning a range of high HIV burden contexts in sub-Saharan Africa: South Africa, Zimbabwe and Kenya, and used to assess strategies for PrEP scale-up in each country, considering cost-effectiveness and population-level impact. Conclusions: PrEP is likely to be of benefit in reducing HIV risk in women across a spectrum of HIV-risk in sub-Saharan Africa, even if reductions in condom use occur. PrEP will be most cost-effective for individuals at great HIV risk, such as female sex workers. However, PrEP has potential to significantly reduce the number of new infections at population-level if made widely available beyond those at highest individual risk, including to women in the general population. Strategies for PrEP scale-up will need to weigh the potential cost-effectiveness and population-level impact of PrEP with the potential for PrEP integration into a wide range of national services and at community level, in order to significantly bring down the costs and improve cost-effectiveness in resource-constrained environments. Static models can be sufficiently robust to inform policy making around the introduction of new HIV prevention interventions in high HIV-burden settings over short-medium time horizon of up to 5 years, where underlying HIV epidemics have reached equilibrium. Over longer timeframes, and in contexts where the underlying HIV epidemics are still evolving (other than over short time horizons of less than a year), static models may under-emphasize situations of programmatic importance and dynamic models will be more appropriate to guide decision making.