This paper argues that the notion of sexual partners per se is insufficient for estimating levels of HIV risk behaviour or changes in HIV risk over time, even though it is a crucial element of most epidemiological models of HIV. The concept of penetrative sexual partner (PSP) is introduced as a considerably more accurate measure of HIV risk. Using data from a longitudinal study of 930 homosexually active men in England and Wales, this paper demonstrates that variation in numbers of PSPs (and thus HIV risk) is not related to variation in gross numbers of sexual partners. Key words: epidemiology; modelling; anal intercourse; gay men; sexual partners; penetrative sexual partners.