Using the nationally representative sample of about 15,000 Add Health respondents in Wave III, the hypothesis is tested that masculinity–femininity in adolescence is correlated with sexual orientation 5 years later and 6 years later: that is, that for adolescent males in 1995 and again in 1996, more feminine males have a higher probability of self-identifying as homosexuals in 2001–02. It is predicted that for adolescent females in 1995 and 1996, more masculine females have a higher probability of self-identifying as homosexuals in 2001–02. Masculinity–femininity is measured by the classical method used by Terman & Miles. For both time periods, the hypothesis was strongly confirmed for males: the more feminine males had several times the probability of being attracted to same-sex partners, several times the probability of having same-sex partners, and several times the probability of self-identifying as homosexuals, compared with more masculine males. For females, no relationship was found at either time period between masculinity and sex of preference. The biological mechanism underlying homosexuality may be different for males and females.