An hierarchical syntaxonomic classification of the vegetation of the Camdebo and Sneeuberg regions of the karoo biome is presented as a second approximation after the earlier work by Acocks (1953). Details on the geomorphology, geology, climate, and early vegetation history of the area are given. The vegetation of the study area was stratified with the aid of Landsat imagery and the community classification was generated using two-way indicator species analysis (Twins pan) which produced ordered phytosociological tables. Tabular comparisons and final sorting of tables are according to the methods and techniques of the ZiirichMontpellier school of phytosociology. Syntaxonomic ranks are defined as five classes, nine orders and seventeen communities. The classes are Grasslands, Karoo Shrublands, Karoo Dwarf Shrublands, Sub-tropical Transitional Thicket, and Riparian Thicket. The distribution of syntaxa corresponds with the steep precipitation gradient experienced in the study area. These vegetation concepts are applied to the description of the flora of the Karoo Nature Reserve and an analysis of the total flora of the reserve is provided. The communities of the pediments, which contain the highest number of endemics, are poorly conserved. I test the validity of the vegetation classification by interpreting the results of an analysis of soils within the hypothesized vegetation units. There is a gradient of increasing Na, silt and pH levels from the Shrublands and Grasslands to the Succulent and Grassy Dwarf Shrublands of the pediments. A qualitative model of the vegetation history during the glacial-interglacial sequence in the Graaff-Reinet region of the eastern Cape is presented. Using a descriptive approach, the distribution patterns of 68 taxa, which are differential species for Karoo Shrublands, Succulent Thicket and Karoo Dwarf Shrublands, are investigated relative to major southern African biomes. The results indicate that a large proportion of the differential species in the phytosociological classification show affinity with Grassland and Savanna Biomes. Three species groups encountered in the Dwarf Shrublands show affinity with the Nama-Karoo biome. The differential species of the Succulent Thicket have a predominantly subtropical distribution. Using an historical approach, the palaeoenvironment of the region during the past 20 000 years is discussed briefly. On the basis of the descriptive and historical perspectives, a qualitative model of vegetation history is presented. The Succulent Thicket may have become established on edaphically favourable sites in the ameliorating conditions of the warmer, wetter Holocene subsequent to the Last Glacial Maximum. The Dwarf Shrubland and Succulent Dwarf Shrubland are depauperate in relation to ccmmunities in other southern African biomes, but the relatively large number of endemics suggests a long history in the region. Their differential species groups occur under arid conditions, accompanied by soils with high base and fertility status. The Dwarf Shrublands may have been more extensive during the drier glacial times on those sites currently occupied by Shrubland. The Shrublands display the expected affInity with the Grassland and Savanna Biomes. The small number of endemics suggest that these communities may have occupied the region in the period since the Last Glacial Maximum. Species with Succulent Karoo Biome affInity are poorly represented. The reliability of using Landsat products to detect and map the vegetation of the region is assessed. The manual classification of Landsat standard products provides a poor reflection of the vegetation of the arid, sparsely-vegetated bottomlands and pediments. The products provide good representation of the boundaries of thicket vegetation, but this uni-temporal approach does not distinguish between floristically different thicket communities. After analyzing digital Landsat data, I suggest that the multi-spectral scanner detects the boundaries of broad soil pedons and geological formations in areas of low vegetative cover. I describe and map the vegetation categories of the region after manual interpretation of six Landsat scenes. This is an effIcient, cost-effective method of mapping vegetation in extensive regions. The mapping units do not reflect the syntaxonomic classification, representing rather an integration of physiographic, pedological, geological and floristic information. With the view to improving the classification of these units, I develop a qualitative model of the natural resources of the region using an expert system