A theoretical framework is proposed on which some hypotheses related to the impact of imperfect vaccines on the evolution of HIV virulence can be tested. For this, a linear increase of risk behaviour with vaccine efficacy is assumed. This is based on the hypothesis that people are prone to relax preventive measures by knowing that they and their partners are vaccinated and that this effect is more intense the more effective the vaccine is known to be. An additional, and perhaps more important hypothesis is related to the theoretical possibility that increased risk behaviour of some vaccinated individuals in partially protected populations could act as a selective pressure toward more virulent HIV strains. Those hypotheses were tested by a mathematical model that considers three different HIV strains competing against each other in a population partially protected by imperfect vaccines of distinct efficacies. Simulations of the model demonstrated that, under the above hypotheses, there is a shift in HIV virulence towards more aggressive strains with increase in vaccine efficacy, associated with a marked reduction in the total amount of transmission and, consequently, in the prevalence of HIV. Potential ways for further testing the theory/model and the implications of the results are discussed.