Using latent class analysis (LCA), we examined patterns of participation in multiple scenes, how sexual risk practices vary by scene, and psychosocial factors associated with these patterns among 470 gay, bisexual, and other men who have sex with men (GBM) recruited from Toronto. We calculated posterior probability of being in a class from participation in nine separate scenes. We used Entropy, the Bayesian information criterion and the Lo-Mendel-Rubin likelihood ratio test to identify the best fit model. Fit indices suggested a four-class solution. Half (50%) of the GBM reported no or minimal participation in any scene, 28% reported participating in the dance club scene, 16% reported participating in the BDSM, bear, and leather scenes, and 6% reported participating in circuit, party and play, and sex party scenes. Compared to GBM who did not participate in scenes, GBM participating in the BDSM-Bear-Leather scene were more likely to be older, white, to report higher sexual self-esteem, and to engage in condomless anal sex; Party and Play scene members were more likely to be of Asian origin, and to use drugs before and during sex, whereas Dance Club scene members were more likely to be younger and to report lower self-esteem but higher hope. LCA allowed us to identify distinct social niches or micro-cultures and factors characterizing these micro-cultures. GBM differ in their risk for HIV and STIs according to characteristics associated with participation in distinct micro-cultures associated with scenes. Tailored interventions may be needed that focus on reducing HIV risk and promoting sexual health in specific contexts such as the BDSM-Bear-Leather and Party and Play.