Habitat selection is a scale-dependent process of paramount importance to the understanding of how species deal with environmental variation. This process has practical implications for wildlife conservation, aiding in the identification of key resources for animals and in the definition of scales relevant to the proposal of practical conservation actions. In this study, we investigated in different spatial scales the habitat selection of the maned three-toed sloth (Bradypus torquatus), a threatened and endemic arboreal folivore of the Atlantic rainforest (vulnerable, sensu IUCN). We radio-tracked and monitored seven sloths for 18 months in landscapes of Southern Bahia, Brazil, the current core region of the species’ distribution. The average values of the home-range estimates were low, but varied considerably among individuals regardless of the estimator (from 0.95 to 27.8ha, MCP method, and from 0.39 to 21.52ha, fixed kernel method). At the landscape scale, the maned sloths preferred early secondary forest and shade cacao plantations, avoided open areas, and occupied late secondary forest as expected compared to its availability. At the home range scale, however, the sloths did not show preference for any forest category, though, again, avoided open areas. At smaller spatial scales, the sloths were highly selective towards forest patches characterized by complex vegetation structures (i.e., areas with a high density of trees, closed and dense canopies), and selected large trees with lianas and bromeliads and also with connected crowns. The high selectivity observed at finer scales appeared to be the result of limited spatial perception and experience due to the species’ characteristic slow mobility. Our results support the notion that maned sloths can effectively occupy (and even select for) disturbed forested habitats. However, we do not know whether sloth populations are viable in landscapes containing only disturbed habitats or low proportions of undisturbed habitats.