ABSTRACTThe faunas of tank bromeliads were sampled over two years in three forest types at different elevations in the Luquillo Experimental Forest, Puerto Rico, and the diversity of their animal communities compared. Bromeliad plants behaved as islands in that, within forests, the species richness and abundance of their animal communities were significantly and positively correlated with increase in plant size. The amount of canopy debris they accumulated was similarly correlated with increase in plant size. Overall diversity was lowest in the dwarf forest, where plants were uniformly small. Animal communities were stable from year to year, and could be characterised for each forest type and for compartments within the plant. They showed a pattern of high dominance, which increased with elevation (Mc‐Naughton index 37, 54, and 73, respectively, for the tabonuco, palo Colorado, and dwarf forest). Alpha‐diversity for sites sampled in each year reflected net primary productivity (NPP) of the forest, declining with increasing elevation when animal abundance measures were used (jackknife estimates of Simpson's diversity index 6.54 & 11.04 [tabonuco], 3.53 & 6.22 [palo Colorado], and 2.75 & 2.17 [dwarf forest]). Species richness over the two years, however, was highest in the intermediate palo Colorado forest (187 species), compared to 146 and 88 in the tabonuco and dwarf forests, respectively. These figures were close to jackknife estimates of maximum species richness. The difference in species richness between tabonuco and palo Colorado forests was significant in one year only. In addition to NPP, other factors, such as litter quality and the structural complexity of the habitat in the palo Colorado forest, may have influenced species richness. The most abundant species in individual plants were also the most widely occurring, confirming known patterns of abundance and distribution in other functional groups. Diversity within bromeliad microcosms at different elevations supported known relationships between diversity, productivity, and habitat complexity along gradients and was not related to differences in the total bromeliad habitat available for colonization.