Correlations between abundance of organisms and their habitat have often been used as a measure of the importance of particular habitat features. However, experimental manipulation of the habitat provides a more unequivocal estimate of its importance. In this study we quantified how fish communities on small patch reefs covaried with changes in benthic cover habitat features. A random sample of small patch reefs was selected and both fish abundance and habitat measures recorded. Naturally occurring patch reefs could be classed into three habitat types based on their benthic cover. Reefs dominated by massive soft corals were the most abundant (50%), followed by those dominated by rock and soft corals in equal proportions (36%), then reefs dominated by branching corals (14%). Fish assemblages differed between the reef types. Communities on soft-coral-dominated and rock/soft-coral-dominated patch reefs formed a continuum of species responses correlated with degree of soft coral cover. In contrast, branching-coral-dominated reefs were occupied by a more discrete set of species. We tested the role of soft corals in contributing to this pattern by experimentally reducing soft coral cover on patch reefs from a baseline level of ~67% to ~33% and ~6%, and monitoring the experiment over 2 years. Contrary to expectations derived from the correlative data, and in contrast with previous manipulations of hard corals, soft-coral disturbance did not generate any corresponding changes in the fish assemblage. This "negative" result indicated that the quality and heterogeneity of habitat generated by soft corals on patch reefs was indistinguishable from equivalent-sized habitat patches formed by bare rock alone. Nevertheless, because soft corals are living organisms they have the potential to generate indirect effects by interacting with other organisms such as hard corals. In the long-term, we hypothesize that biotic interactions between habitat forming organisms might affect composition of fish assemblages on patch reefs.