The aim of this study was to document the composition and distribution of deep-water fishes associated with a submarine canyon-valley feature. A work-class Remotely Operated Vehicle (ROV) fitted with stereo-video cameras was used to record fish abundance and assemblage composition along transects at water depths between 300 and 900 metres. Three areas (A, B, C) were sampled along a submarine canyon-valley feature on the continental slope of tropical north-western Australia. Water conductivity/salinity, temperature, and depth were also collected using an ROV mounted Conductivity Temperature and Depth (CTD) instrument. Multivariate analyses were used to investigate fish assemblage composition, and species distribution models were fitted using boosted regression trees. These models were used to generate predictive maps of the occurrence of four abundant taxa over the survey areas. CTD data identified three water masses, tropical surface water, South Indian Central Water (centred ∼200 m depth), and a lower salinity Antarctic Intermediate Water (AAIW) ∼550 m depth. Distinct fish assemblages were found among areas and between canyon-valley and non-canyon habitats. The canyon-valley habitats supported more fish and taxa than non-canyon habitats. The fish assemblages of the deeper location (∼700–900 m, Area A) were different to that of the shallower locations (∼400–700 m, Areas B and C). Deep-water habitats were characterised by a Paraliparis (snail fish) species, while shallower habitats were characterised by the family Macrouridae (rat tails). Species distribution models highlighted the fine-scale environmental niche associations of the four most abundant taxa. The survey area had a high diversity of fish taxa and was dominated by the family Macrouridae. The deepest habitat had a different fish fauna to the shallower areas. This faunal break can be attributed to the influence of AAIW. ROVs provide a platform on which multiple instruments can be mounted and complementary streams of data collected simultaneously. By surveying fish in situ along transects of defined dimensions it is possible to produce species distribution models that will facilitate a greater insight into the ecology of deep-water marine systems.