It has previously been demonstrated that the distribution of sexual partner change rates may not be sufficient behavioral information to predict the spread of AIDS. A more adequate description of the sexual network through which AIDS may spread will require information on who mixes with whom. Previous models indicated that a highly "assortive" pattern of sexual mixing, in which people of high partner change rates nearly always partner similar people, would result in a relatively rapid epidemic but one that would ultimately be rather small as it would be largely confined to those with high partner change rates. However, in earlier models it was necessary to arbitrarily vary either the partnering patterns or the distribution of partner change rates as the epidemic progressed in order to ensure consistency in the partnering patterns. Consequently, the "pure" effects of assortiveness on the size of the epidemic could not be assessed. In this paper a new model is described in which consistency in partnering patterns is maintained as the epidemic selectively depletes those with higher partner change rates. However, this model maintains consistency without arbitrary adjustments to either the assortive nature of partnering or important features of the distribution of the partner change rate. It is then possible to test the effects of assortiveness on the outcome of the epidemic without important confounding effects from the mathematical device used to maintain consistency in partnering. A practical sensitivity test on the impact of assortiveness on model predictions shows that, contrary to expectations, increasing assortiveness can, under certain circumstances, lead to larger epidemics.