Species' traits have been widely championed as the key to predicting which species are most threatened by habitat loss, yet previous work has failed to detect trends that are consistent enough to guide large-scale conservation and management. Here we explore whether traits and environmental variables predict species sensitivity to habitat loss across two datasets generated by independent avifaunal studies in the Atlantic Forest of Brazil, both of which detected a similar assemblage of species, and similar species-specific responses to habitat change, across an overlapping sample of sites. Specifically, we tested whether 25 distributional, climatic, ecological, behavioral and morphological variables predict sensitivity to habitat loss among 196 bird species, both within and across studies, and when data were analysed as occurrence or abundance. We found that 4-9 variables showed high explanatory power within a single study or dataset, but none performed as strong predictors across all datasets. Our results demonstrate that the use of species traits to predict sensitivity to anthropogenic habitat loss can produce predictions that are species- and site-specific and not scalable to whole regions or biomes, and thus should be used with caution. This article is protected by copyright. All rights reserved.