The patch structure of the grass sward at Kroomie (26°25'38"E 33°48'30"S) in a semi-arid savanna in the Eastern Cape, South Africa, was investigated. The study was conducted on long term grazing trials on five treatments varying in stocking rate (SR; recommended (low) and 1.5 x recommended (high), grazing system (continuous and rotational), and animal type (cattle and sheep). The treatments studied were CR (cattle, rotational stocking, low SR), CC (cattle, continuous stocking, low SR), CH (cattle, rotational stocking, high SR), SC (sheep, continuous stocking, low SR), and SR (sheep, rotational stocking, low SR). Rainfall during the two years of the study (1997/98 and 1998/99) was slightly below the mean average rainfall of the area (66 and 84% of the mean of 519 mm, respectively). Analysis of sward height data using Maximum Likelihood Estimation reflected a bimodal height structure in all treatments. Due to a high overlap of the two distributions in some cases, however, the height at which to separate patches (short grass) from non-patches (tall grass) could not be determined. Canonical Correspondence Analysis (CCA) was used to relate species composition to sward height. It emerged that there are two distinct grass communities at Kroomie, and these are associated with sward height (i.e. patches and non-patches). The interface (in cm) between these two communities, as determined using Two Way Indicator Species Analysis (TWINS PAN) was 6 cm, and this value was subsequently used to discriminate between patches and non-patches. Sward structure was affected by treatments. Animals (cattle and sheep) stocked rotationally at low SR's grazed less than a third of the total area, and this grazing was concentrated primarily in small patches (< 6 m; length is used as a linear indicator of patch size). Animals stocked continuously at low SR's grazed approximately half the area, in small and large (up to 40 m) patches. Animals in the CH treatment grazed approximately two-thirds of the area, in both small and large patches. There was an inverse relation between the size of patches and the size of non-patches, as expected. Nine common grass species were related to sward height. Digiteria eriantha, Eragrostis racemosa, Eustachys paspaloides, and Microchloa caffra were associated with short swards, while Cymbopogon pluronodis, Eragrostis chloromelas, and Sporobolus fimbriatus were associated with tall swards. Themeda triandra (themeda), the most abundant grass at Kroomie, was principally associated with tall swards, but was present at all sward heights. Applying CCA demonstrated a considerable difference between the species composition of patches and of non-patches. There was also a difference in composition between treatments, although these were not as pronounced. Patches reflected a higher species diversity than non-patches. There was a significant (P<0.05) effect of treatment, and of an interaction of treatment by sward structure (i.e. patches and non-patches), on the density of themeda plants. The density of themeda plants was positively correlated with patch size, which suggested that themeda plants that have been grazed may suffer fatal competition from ungrazed neighbours. Anecdotal evidence suggested that patches are stable over the medium term, and that non-patches that are grazed during a drought return to a non-patch structure after rainfall. There was no evidence to support the contention that rotational stocking reduced patch-selective grazing, nor that the species composition of rotationally stocked treatments was better than continuously stocked treatments. / Thesis (M.Sc.)-University of Natal, Pietermaritzburg, 2003.