1. Underwater visual census (UVC) using SCUBA is a commonly used method for assessing reef fish communities. Evidence suggests, however, that fish avoid divers due to the sound of bubbles produced by open-circuit SCUBA, and avoidance behaviour is more pronounced as fishing pressure increases. Despite the potential for producing biased counts and conclusions, these behavioural effects have rarely been quantified, especially when assessing the effectiveness of marine protected areas (MPAs). 2. To test the magnitude of avoidance behaviour, we surveyed fish populations inside and outside two MPAs in Guam, using two diving systems: standard open-circuit (OC) SCUBA and a closed-circuit rebreather (CCR) that produces no bubbles. Data were collected using a diver-operated stereo-video system (stereo-DOV), which provided counts of relative abundance, measures of fish length and the minimum approach distance of the diver to a fish. 3. Inside MPAs, fish surveys conducted with CCR recorded similar community metrics to fish surveys conducted with conventional OC SCUBA. In contrast, outside the MPAs, the bubble-free diving system recorded 48% more species and up to 260% greater fish abundance. These differences reflected the ability of a diver wearing the silent CCR unit to sample the larger, most heavily targeted species that are shy of divers in fished areas. This difference was also large enough to change some results from ‘reject’ to ‘accept’ the null hypothesis of ‘no significant differences exist between fished and protected areas’. 4. The use of CCR for fish surveys clearly minimizes behavioural biases associated with fish avoiding open-circuit SCUBA divers. We recommend the use of this bubble-free diving system for surveys assessing reef fish populations, especially in areas where fish are heavily targeted by spearfishing. If fish behaviour is not accounted for, surveys using SCUBA could result in erroneous conclusions when comparing fished and protected areas. While the behaviour of fish towards divers is rarely mentioned in conclusions from studies using UVC, it is an important source of bias that should be acknowledged and minimized where possible.