Most current candidate HIV vaccines seem to produce little protection against infection, but reduce viral load and slow the decline in CD4 lymphocyte numbers. Such disease-modifying vaccines could potentially provide important population-level benefits by reducing transmission, but could possibly also increase transmission. We address the following question: could disease-modifying HIV vaccines cause population-level perversity (ie, increase epidemic severity)? By analysing a mathematical model and defining a new quantity-the fitness ratio-we show that disease-modifying vaccines that provide only a low degree of protection against infection and/or generate high fitness ratios will have a high probability of making the epidemic worse. However, we show that if disease-modifying vaccines cause a 1.5 log(10) reduction in viral load (or greater) then perversity cannot occur (assuming risk behaviour does not increase). Finally, we determine threshold surfaces for risk behaviour change that determine the boundary between beneficial and perverse outcomes; the threshold surfaces are determined by the fitness ratio, the proportion of the population that are "successfully vaccinated", and the degree of change of risk behaviour in unvaccinated infected individuals. We discuss the implications of our results for designing optimal vaccination control strategies.