Anthropogenic activities and greater demands for marine natural resources has led to increases in the spatial extent and duration of pressures on marine ecosystems. Remotely operated vehicles (ROVs) offer a robust survey tool for quantifying these pressures and tracking the success of management intervention while at a range of depths, including those inaccessible to most SCUBA diver-based survey methods (~>30 m). As the strengths, limitations, and biases of ROVs for visually monitoring fish assemblages remain unclear, this review aims to evaluate ROVs as a survey technique and to suggest optimal sampling strategies for use in typical ROV-based studies. Using the search engines Scopus™ and Google Scholar™, 119 publications were identified that used ROVs for visual surveys of fish assemblages. While the sampling strategies and sampling metrics used to annotate the imagery in these publications varied considerably, the total abundance of fish recorded over strip transects of varying dimensions was the most common sampling design. The choice of ROV system appears to be a strong indicator of both the types of surveys available to studies and the success of ROV deployments. For instance, larger, more powerful working-class systems can complete longer and more complex designs (e.g., swath, cloverleaf, and polygonal transects) at greater depths, whereas observation-class systems are less expensive and easier to deploy, but are more susceptible to delays or cancelations of deployments. In more severe sea state conditions, radial transects, or strip transects that employ live-boating or a weight to anchor the tether to the seafloor, can be used to improve the performance of observation-class systems. As these systems often employ shorter tethers, radial transects can also be used to maximize sampling area at greater depths and on large vessels that may rotate substantially while anchored. For highly mobile species, and in survey designs where individuals are likely to be recounted (e.g., transects along oil and gas pipelines), relative abundance (MaxN) may be a more robust sampling metric. By identifying subtle, yet important, differences in the application of ROVs as a tool for visually surveying deep-water marine ecosystems, we identified key areas for improvement for best practice for future studies.