Our objective was to examine the representativeness of samples of homosexuals, bisexuals, gays, and lesbians obtained for public health research. We identified journal articles cited in Medline and published between 1990 and 1992 that sampled individuals and classified them as homosexual, bisexual, gay, and/or lesbian. Information was abstracted from these articles to evaluate four components of sample selection affecting the representativeness of samples: (1) how the population is conceptually defined, (2) how the sampled population is operationally identified, (3) the setting from which samples are selected, and (4) the use of probability sampling to select subjects. We identified 152 public health articles published between 1990 and 1992 that sampled homosexuals, bisexuals, gays, and/or lesbians. We found articles (1) rarely (4/152) conceptually defined the population they were sampling, (2) used a range of incomparable methods to identify and select subjects, (3) sampled from settings representative of dramatically different populations, and (4) rarely (3/152) used probability sampling. Overall, we find methods used to identify homosexuals, bisexuals, gays, and lesbians for public health research produce samples representative of different and sometimes unidentifiable populations. To understand these populations from a public health perspective, it is imperative that steps be taken by researchers to standardize population definitions and sampling methodologies.