The fact that species vary in their vulnerability to extinction is well documented, but the reasons for these differences remain poorly understood. Why should some species/families/guilds decline rapidly with increasing anthropogenic disturbance, while others either tolerate or proliferate in disturbed habitats? We investigated the bird species composition in 31 primary forest patches of varying size in a region of the Amazonian 'Arc of Deforestation' and assessed which species life-history traits predisposed individual species to extinction. Medium-sized non-flocking canopy frugivores/ominvores of low primary forest dependence were least likely to go extinct in small patches, while small-bodied flock-following primary-forest-dependent terrestrial insectivores were most fragmentation sensitive. We found highly idiosyncratic relationships between the minimum size of forest patches occupied by different species and their territory size requirements estimated based on other Amazonian studies. This suggests that avian assemblages in forest fragments primarily comprise species that either have good dispersal abilities or are highly tolerant to the non-forest matrix, rather than those whose minimum spatial requirements can be met by the size of available forest fragments.