The degree to which species partition resources often depends on the spatial scale of the study. To investigate this, we examined the distribution and abundance patterns of 3 relatively rare pygmy angelfishes (family Pomacanthidae: Centropyge) among reef locations, depths and microhabitats at Lizard Island on the Great Barrier Reef. A strong association among species and between species and their habitat was found at some scales, but not others. On a broad scale, the abundances of the 3 species were highest at the same 4 sites (kilometres apart). These sites were all located at headlands, suggesting that the patterns of abundance may be in response to the topographic features of the island at this scale. Differences among species occurred at intermediate scales (10s to 100s of metres), where species were associated with different depth zones or reef locations. Centropyge vroliki occurred shallower than C. bicolor, while the depth distribution of C. bispinosa overlapped with both of these species. Laterally along the fringing reef, C. vroliki were more abundant in areas where both C. bicolor and C. bispinosa were also abundant, but these 2 latter species were not correlated with each other. The proportions of substratum types present in home patches differed among species, to some extent reflecting the benthic composition of the reef area where each occurred. However on a fine scale (metres), all 3 species appeared to use the same substratum type, which consisted of dead branching coral covered in algae, and occasionally formed multi-species groups. Species abundances were not correlated with this commonly used microhabitat, but rather the availability of the substrata characterizing the reef areas in which they were most abundant. Here, although Centropyge species use the same type of microhabitat, they may be an example of species that partition space on the basis of non-preferred resources. For example, all 3 species used home patches containing high proportions of overgrown corals; however where this habitat was not available, C. bicolor used sand and rubble habitat at the reef base, while C. vroliki used coral habitat on the reef crest. Most importantly, this study emphasizes that a multi-scale approach is necessary to determine appropriate scales for examining species associations and resource partitioning in reef fishes.