Monitoring fish and underwater habitats, particularly in and around marine protected areas (MPAs) requires non-destructive observation methods. This is generally achieved by divers conducting underwater visual censuses (UVC), but video-based techniques are now being used more often to observe underwater macrofauna and habitats. A comparison of these two techniques is relevant with the development of high-definition (HD) video, which constitutes a substantial improvement over previously available video resolutions at limited extra cost. We conducted a paired observation experiment involving both HD video and UVC in an MPA located in the New Caledonian lagoon, which is a highly diversified coral reef ecosystem. We compared three techniques for counting fish along 50 m x 4 m delineated strip transects: UVC and two video techniques in which the diver used either a straight trajectory (I-type transect) or a browsing one (S-type transect). The results showed that the proportion of fish that were not identified up to the species level did not exceed 3.3% in video observations versus 1.7% in UVC. The abundance and species richness were larger in UVC than in videos, and S-type transects detected more individuals and species than I-type transects. The average abundance and species richness observed by UVC were 1094 individuals and 69.7 species per transect respectively. In comparison with UVC. I-type and S-type video transects detected on average 56% and 61% of the abundance and 85% and 77% of the species richness seen by UVC respectively. Our results showed that, in comparison to UVC data recorded in situ, the post field analysis of HD video images provided representative observations of fish abundance and species diversity, although fewer species and individuals were detected. The advantages and shortcomings of each observation technique for monitoring fish assemblages, particularly in an MPA are discussed. HD video appears to be a cost-effective technique in terms of the human resources and time needed for field implementation. Overall, this study suggests that HD video-based techniques constitute an interesting complement to UVC, or an alternative when these cannot be implemented. (C) 2010 Elsevier B.V. All rights reserved.