The ongoing loss of large trees and densification of shrubs are two prevalent processes that take place in African savannas, with profound consequences for their structure and function. We evaluated herbivore impacts on savanna woody communities using a long-term exclosure experiment in the Kruger National Park, South Africa, with three treatments: the exclusion of large mammals only (i.e. elephant and giraffe), exclusion of all herbivores larger than a hare, and areas open to all herbivores. We asked three questions: (1) How did variable exclusion of herbivores affect woody density and structure across the catena (i.e. riparian, sodic and crest vegetation)? (2) Did the exclusion of herbivores result in unique woody species composition? (3) Did herbivore exclusion result in a higher proportion of palatable species? After 17 years, we found that herbivores mainly affected the heights and densities of existing species, rather than leading to turnover of woody species assemblages. Although densities of individuals increased in the full exclosure (350 ha −1 ), the change was more moderate than expected. By contrast, mixed mega-and meso-herbivores decreased the number of trees and shrubs (decreases of 780 ha −1 ) via a variety of physical impacts. Meso-herbivores alone, on the other hand, had less impact on individual density (i.e. no change), but limited average height growth and canopy dimensions in certain habitat types. Where elephants are present, they are effective at reducing the density of woody stems to the point of counteracting woody encroachment, but at the same time are actively preventing the persistence of large trees (>5 m) as well as preventing trees from recruiting to larger size classes. However, the lack of massive recruitment and woody cover increases with elephant exclusion, especially for more preferred species, suggests that factors beyond elephants, such as dispersal limitation, seed predation, and drought, are also acting upon species.