Although underwater visual census (UVC) is the most frequently used technique for quantifying reef fish assemblages, remote video analysis has been gaining attention as a potential alternative. In the South Atlantic Ocean, Millepora spp. (class Hydrozoa) are the only branching coral species; however, little is known about the ecological role that they play for reef fish communities. We compared these two observation methods (remote video and UVC) to estimate reef fish abundance and species richness associated with colonies of the firecoral Millepora alcicornis at Tamandar e Reefs, Northeast Brazil. Additionally, the two different techniques were used to compare species behaviour in association with fire-corals in order to examine the biases associated with each technique and provide useful information for behavioural ecologists studying fishcoral associations. There were no differences in reef fish abundance or species richness sampled by remote video or UVC. However, a significant difference in the behaviour of associated fish was recorded between the two methods. In the presence of a diver carrying out a UVC, fish were observed spending more time sheltered amongst the coral branches compared with passively swimming on coral colonies with the remote video technique. Specifically, on the remote video recordings agonistic interactions between fish and passive swimming accounted for 33.3% and 22.2% of the census time, respectively. By comparison, when observed by a diver fish spent 34.8% of their time sheltering amongst the coral branches. We demonstrate that both techniques are similarly effective for recording fish abundance and species richness associated with firecorals. However, differences were observed in the ability of each method to detect the behaviour of coral-associated fishes. Our findings show that behavioural ecologists studying complex fish-coral associations need to ensure that their aims are clearly defined and that they choose the most appropriate technique for their study in order to minimize methodological biases.