The aim of the present study was to determine whether boat-based deployment of remote underwater video cameras is effective for surveying fish assemblages in the deepest reaches of two large tropical rivers in north-eastern Australia. In addition, we compared fish assemblages recorded on baited versus unbaited cameras, and evaluated the sampling effort (duration of recording) required to estimate fish assemblages using remote underwater videos. We found that fish assemblages differed according to the depth, with statistically significant differences largely attributable to the prevalence of small-bodied species (<10-cm total length, TL), such as Ambassis sp., Melanotaenia sp. and Pseudomugil signifer recorded in shallow (0.4–2.0 m) and intermediate (2.1–4.9 m) depths, and larger-bodied fish species (>10 cm TL), such as Lutjanus argentimaculatus, Mesopristes argenteus and Caranx sexfasciatus, in deep water (>5.0 m). Estimates of fish assemblage attributes generally stabilised after 60 min recording duration, suggesting that interrogation of video footage beyond this duration may not be cost-effective. We conclude that depth is an important consideration when surveying large and deep river fish assemblages and that where water clarity is favourable, underwater video provides one of the means by which an assemblage can be investigated across the entire depth profile.