Question: Can a skills training intervention reduce sexual risk behaviours in homeless men who have mental illness? Design: Randomised controlled trial with 6 and 18 months follow up. Setting; 200 bed municipal men's shelter in New York City, New York, USA. Patients: 97 men attending a psychiatric outreach programme. 59 of the men were sexually active at baseline (defined as having had vaginal or oral sex in the past 6 mo); most were >/= 35 years of age (58%), African-American (58%) or Latino (35%), had not completed high school (56% and had psychiatric diagnoses of schizophrenia or schizoaffective disorder (57%) or major or bipolar depression (27%). For sexually active men, follow up was 100% at 6 months and 95% at 18 months. Intervention: 52 men were allocated to the Sex, Games, and Videotapes (SexG) intervention, which involved 15 interactive sessions (2 d/wk for 8 wks) facilitated by a mental health professional and a paraprofessional. Activities included storytelling, competitive games, acting scenes from their daily lives, and practising the decision making skills required for consistent condom use. 45 men were allocated to the control intervention which involved 2 sessions on HIV infection and other sexually transmitted diseases and basic instruction on condom use. Main outcome measures: Main outcome was a sexual risk index (the Vaginal Episode Equivalent [VEE]) based on a semistructured interview given by interviewers blinded to the participant's group assignment. The index assigned scores for each unprotected anal, vaginal, and oral sexual episode. Other outcomes included episodes of sex, episodes with condom use, and high risk episodes (multiple partners and unprotected anal or vaginal sex). Main results: Intention to treat analysis focused on the subgroup of men who were sexually active at baseline (n = 33 in the SexG group and n = 26 in the control group). At 6 months, sexually active men in the SexG group had a lower mean VEE index than sexually active men in the control group (1.0 v 3.1, p = 0.01). Although the total number of episodes of sex were similar for the 2 groups (261 v 247), the SexG group had a higher proportion of episodes with condom use than the control group (73% v 52%, p = 0.04), and a lower proportion of high risk episodes (24% v 58%, p=0.01). The between group difference in mean VEE scores decreased over time from 2.1 at 6 months to 1.6 at 18 months. Among the 38 men who were not sexually active at baseline (n = 19 in each group), the SexG group did not differ from the control group for number who were sexually active (11 v 9) or had unprotected sex (5 v 6) at 18 months. Conclusion: A skills training intervention reduced high risk sexual behaviours in sexually active homeless men who had mental illness. [ABSTRACT FROM AUTHOR]