Global trends of transformation and loss of wetlands to other land uses has deleterious effects on surrounding ecosystems, and there is a resultant increasing need for improved mapping of wetlands. This is because wetland conservation and management depends on accurate spatial representation of these systems. Current approaches to mapping wetlands through the classification of satellite imagery typically under-represent actual wetland area, and the importance of ancillary data in improving the accuracy in mapping wetlands is recognized. This study uses likelihood estimates of wetland occurrence in KwaZulu-Natal (KZN), South Africa, using a number of environmental surrogate predictors (such as slope, rainfall, soil properties etc.). Using statistical information from a set of mutually independent environmental variables in known wetland areas, conditional probabilities were derived through a Bayesian network (BN) from which a raster layer of wetland probability was created. The layer represents the likelihood of wetlands occurring in a specific area according to the statistical conditional probability of the wetland determinants. Probability values of 80% and greater also accounted for approximately 6% of the KZN area (5 520 km²), which is substantially more than the previously documented wetland area in KZN (4% of the KZN area or 4 200 km²). Using an independent test dataset, Receiver Operating Characteristic (ROC) curves with the Area Under Curve (AUC) analysis verified that the final model output predicted wetland area well (AUC 0.853). Based on visual comparisons between the probability layer and ground verified wetland systems, it was shown that high wetland probability areas in the final output correlated well with previously highlighted major wetland and wetland-rich areas in KZN. Assessment of the final probability values indicated that the higher the probability values, the higher the accuracy in predicting wetland occurrence in a landscape setting, irrespective of the wetland area. It was concluded that the layer derived from predictor layers in a BN has the potential to improve the accuracy of the KZN wetland layer by serving as valuable ancillary data. Application of the final probability layer could extend into the development of updated spatial freshwater conservation plans, potentially predicting the historical wetland extents, and as input into the land cover classification process. Keywords: ancillary data, Bayesian network, GIS, modelling, probability, wetland mapping. / Thesis (M.Sc.)-University of KwaZulu-Natal, Pietermaritzburg, 2014.