Understanding ecological traits of species associated with vulnerability to habitat fragmentation is important for the development of effective conservation efforts. However, few studies have examined ecological correlates of vulnerability to fragmentation among tropical small mammals at fine spatial scales. Here we use random forests and regression tree models to identify ecological correlates of vulnerability to fragmentation among terrestrial small mammal species across 2 networks of forest fragments in the East and West Usambara Mountains, Tanzania. Small mammal species vary considerably in their vulnerability to fragmentation as assessed by species occupancy and change in abundance. Random forests models indicate that matrix use is the highest-ranked predictor of vulnerability to fragmentation with small mammal species that do not use or use few matrix habitat types, being more vulnerable to fragmentation than species that use all adjacent matrix habitat types. Our results emphasize the importance of local-scale approaches for designing conservation strategies for Afrotropical small mammal species.