Commercial livestock production offers one of the main opportunities for mainstreaming of biodiversity conservation in the grassland biome of South Africa. Grazing management is expected to influence success. With the uses of three long-term grazing trials, effects of stocking rate and cattle-to-sheep ratio on the plant composition and diversity of Highland Sourveld grassland in KwaZulu-Natal were examined. Plant diversity was sampled with the use of modified Whittaker plots. Canonical correspondence analysis was used to test the effects of treatments on compositional variation, and general linear models were used to test individual species' responses. In a biennial rotation, burned/grazed plots supported lower species richness of forbs and all plants than unburned/ungrazed plots, attributed to the impact of grazing during the season of occupation. A high stocking rate resulted in a long-term decrease of forb richness in one experiment, but an increase in another. An increasing proportion of sheep to cattle resulted in a long-term decrease of the richness of forbs and of total species richness. The three trials identified nongrass species that behaved as increasers or decreasers in response to an increase in stocking rate, and a set of species that behaved as decreasers in response to an increasing proportion of sheep to cattle. Constraints on using long-term trials for identifying the effects of livestock management on plant diversity include lack of baseline data, limited replication, pre-experimental impacts on the study site, and the difficulty of assessing uncommon species.