Deterministic simulation models are used to show that HIV transmission dynamics in homosexual populations can be strongly affected by sexual partner selectiveness. The type of selectiveness or biased mixing examined is where individuals with similar new partnership formation rates are more likely to form a pair than would be expected by chance. The effect of such selectiveness could be strong even when the total number and distribution of new sexual partnerships and sex acts remains constant. This means that in order to predict the future course of HIV transmission and identify the populations at highest risk, we must have information not only on the frequency of new sexual partnerships and types of sex acts, but also on who has sex with whom. Given high sexual partner selectiveness, some groups of homosexuals with low rates of sex and new sex partners would take many decades before a single introduction would generate an epidemic. Epidemics in these groups can be markedly accelerated by only modest contact with higher risk groups. Even in very low activity groups, which if isolated would have no epidemic, an important proportion of their members can be infected when they are not selective. The relative risks of AIDS in groups making high numbers of new sexual partnerships compared to groups making low numbers are markedly affected by sexual partner selectiveness. The models developed were examined using information collected in 1984 from the Coping and Change Study in collaboration with the Chicago Multicenter AIDS Cohort Study. This population was divided into activity groups by the rate at which individuals established new sexual partnerships and then a structure of new sexual partnerships between these activity groups was defined consistent with available data. Even without introducing any behavior change in the models, the proportion of the homosexual population infected was seen to level off temporarily at around 50% after several years as a consequence of saturation in the high risk groups when lower risk groups were not yet being consumed by the epidemic process. Most sex in the study population is casual or anonymous. The sample selection procedures were biased toward individuals who engage in such sex. There is, however, a large group of individuals that avoids the casual and anonymous sex scene. Simulated epidemics indicated that under a variety of conditions, the population that does not engage in casual and anonymous sex could experience an epidemic of HIV infection quite some time after the epidemic in the population that does has waned.(ABSTRACT TRUNCATED AT 400 WORDS)