Repeated behavioural surveillance should sample all epidemiologically relevant subgroups to provide a complete picture of trends in HIV risk behaviours. Web-based recruitment has been mooted but little empirical data exist on country experiences. We describe who is omitted from three rounds of a conventional offline-only surveillance programme among men who have sex with men (MSM) 2006-2011, but recruited subsequently on Internet dating sites, and the implications of this for understanding trends. The latter were younger, less gay identified and less gay community attached. Importantly, they reported different partnering patterns, lower condom use with casual and fuckbuddy-type male partners, and lower rates of HIV testing, compared to MSM routinely captured in offline surveillance. The replacement of offline socio-sexual activity by the Internet among many MSM means that current venue-based surveillance systems may underestimate risk behaviours, overlook trends among unsampled online MSM, and misinterpret trends observed in sampled MSM due to "sample drift" of most-at-risk MSM.