Intuitively, termites would seem to be a very suitable group to illustrate effects of ecosystem fragmentation. Being detritivores, they do not control directly the rate at which their resources are available, nor do they restrict the ability of the resources to regenerate. Consequently, termites do not mask the ecosystem depletion caused by fragmentation. With this in mind, we compared the communities of termites in undisturbed Amazonian forest with those of two isolated fragments nearby, aiming to show that the differences observed may have resulted from habitat fragmentation. Dissimilarities between communities in the undisturbed forest suggest natural patchiness in their distribution, which could lead to misinterpretation of the effects of fragmentation. Continuous forest had higher species richness and fewer rare species than the fragments. Guild structure in the forest was biased towards soil-feeding termites, which are subterranean and soft bodied, and therefore more sensitive to variation in microclimate. In the fragments, litterfeeders and species intermediate between soil-feeding and wood-feeding types were numerically more important. Habitats in the forest were more equally used than in the fragments, suggesting habitat unsuitability increased with fragmentation. It is suggested that the community composition of the fragments is a result of the intrinsic patchiness of the original forest and deterministic and stochastic extinctions caused by fragmentation. The need for manipulative experiments to test such ideas is discussed.