Gay men's stereotypes about who is HIV-infected were investigated. Young uninfected (n=62), older uninfected (n=61), and infected (n=65) gay men read brief descriptions of men they did not know and estimated the likelihood that they were infected. Each description highlighted one characteristic of the man described. There were 3 versions of each sketch; the versions highlighting preferred sexual practice, for example, described the man as either preferring insertive anal intercourse, preferring receptive anal intercourse, or liking both equally. Results were largely the same for the 3 sample groups. For 6 of the 9 characteristics investigated-preferred haunts, preferred sexual practice, dress code, access to gay venues, occupation, and sexual orientation-significantly different estimates were given for the different versions. Results are discussed in relation to how AIDS education might counter the use by gay men of stereotypes to infer whether a given sex partner is infected. It is suggested that these stereotypes are likely to be present 'on line' (during actual sexual encounters), rather than 'off line' tin the cold light of day), thereby complicating the task of AIDS educators.