Persistence of forest-dependent species in fragmented landscapes strongly relies on sufficient dispersal between patches, making it important to understand how animal movements are affected by the intervening matrix. Movements can be influenced through selection or avoidance of land cover based on their perceived suitability for foraging or providing cover. The composition and configuration of the matrix will, therefore, most likely be an important factor to consider when estimating connectivity between patches. To address this, we performed translocation experiments to understand how forest birds used different land cover types in a fine-grained matrix of a fragmented Afromontane biodiversity hot spot (Taita Hills, Kenya). Our results revealed that use of land cover types for both the forest specialist Cabanis's greenbul Phyllastrephus cabanisi and for the forest generalist white-starred robin Pogonocichla stellata was disproportional to their availability. However, this effect was influenced by matrix configuration; in patchy matrices, land cover selection was more pronounced compared with more uniform matrices, especially for the forest specialist. At the scale of movement steps, risk avoidance seemed to be a strong factor in the route decisions for both species. Observed steps contained on average lower proportions of open land cover and did less frequently intersect built-up areas than expected. P. stellata did not differentiate between the alternative land cover types, whereas P. cabanisi preferred steps that contained more indigenous forest. The observed negative relationship between degree of forest dependency and matrix permeability implies that for members of the Taita bird community, which are even more dependent on intact forest habitat (i.e. the critically endangered Taita thrush Turdus helleri), current permeability of the matrix may be even lower. Matrix restoration to improve connectivity may, therefore, be a crucial instrument for the long-term survival of forest-dependent species in these fragmented Afromontane landscapes.