SUMMARYThe loss of natural habitats is a major threat to biodiversity, and protected area designation is one of the standard responses to this threat. However, greater understanding of the drivers of habitat loss and of the circumstances under which protected areas succeed or fail is still needed. We use visual assessment of satellite images to quantify land-cover change over periods of up to 30 years in and around a matched sample of protected and unprotected Important Bird and Biodiversity Areas (IBAs) in Africa. We modelled the annual survival of forests and other natural land covers as a function of a range of environmental and anthropic predictors of plausible drivers. The best-supported model indicated that survival rates of natural land cover were highest in steeper areas, at higher altitudes, in areas with lower human population densities and in areas where the cover of natural habitats was already higher at the start of the period. Survival rates of natural land cover in protected areas were, on average, around twice those in unprotected areas, but the differences between them varied along different environmental gradients. The overall survival rates of both protected and unprotected forests were significantly lower than those of other natural land-cover types, but the net benefit of protection, in terms of the absolute difference in rates of loss between protected and unprotected sites, was higher in forests. Interaction terms indicated that as slope and altitude increased, the natural protection offered by topography increasingly nullified the additional benefits of legislative protection. Furthermore, protected area designation offered reduced additional benefits to the survival of natural land cover in areas where rates of conversion were higher at the start of the observation period. Variation in the impacts of protected area status along different environmental gradients indicates that targets to improve the world's protected area network, such as Aichi Target 11 of the Convention on Biological Diversity, need to look beyond simple area-based metrics. Our methods and results contribute to the development of a protocol for prioritizing places where protection is likely to have the greatest effect.