Aquatic environmental DNA (eDNA) surveys have emerged as an alternative method for monitoring complex and vast marine ecosystems. One-to-one comparisons between existing survey techniques and eDNA approaches are essential to determine biases associated with this novel methodology. To date, such direct comparative studies have been scarce in the context of marine eDNA surveys. In this study, we conducted simultaneous baited remote underwater video (BRUV) and eDNA surveys to describe the fish community in Paterson Inlet, Stewart Island/Rakiura, New Zealand. BRUV detected three distinct families of bony fish (Actinopterygii) and four families of cartilaginous fish (Chondrichthyes). Three different eDNA assays, detected 32 (MiFish-U), 42 (MiFish-E), and 23 (16S-Fish) families, spanning the classes of Actinopterygii, Chondrichthyes, Hyperoartia, Mammalia, and Aves. Our direct comparison identified the need for (i) increased sampling, (ii) spatial pooling, and (iii) multiple targeted eDNA assays, to achieve similar detection rates of a given species in eDNA and BRUV monitoring. Diversity, ordination, and indicator species analyses identified distinct eDNA signals between different habitats in our relatively small sampling area, showcasing the high spatial resolution of eDNA approaches in marine habitats. Our results provide valuable insights into the potential biases associated with eDNA monitoring, as well as highlight the power of eDNA for detecting a broad range of taxa beyond traditional observational approaches, including terrestrial, invasive and migratory organisms.