Marine reserves are being widely implemented as a tool for fisheries management and biodiversity conservation. Although the siting of marine reserves often includes a surrogate measure of diversity, the precision of these measures is rarely tested. To create the marine park at Lord Howe Island, habitat was used as a surrogate for community diversity. The aims of this study were to test the precision of habitat in predicting reef fish assemblage structure, and to investigate changes in precision when varying resolutions of baseline habitat data were available. Visual counts of reef fish species and habitat surveys were conducted at 31 sites. Overall, the variations in fish assemblage among sites were moderately correlated with habitat variations, while fish assemblages were weakly spatially auto-correlated, strongly affecting sites within a proximity of 1 km. The ability of habitat classes to predict reef fish assemblage structure was dependent on the quality and quantity of baseline data. Differences in assemblage structure were found among habitat classes derived from detailed high-resolution data, but not among habitat classes defined from low-resolution data. This study highlights the need for accurate in situ ecological information to establish precise habitat surrogates to more effectively site marine reserves.