Information on distribution and relative abundance of species is integral to sustainable management, especially if they are to be harvested for subsistence or commerce. In northern Australia, natural landscapes are vast, centers of population few, access is difficult, and Aboriginal resource centers and communities have limited funds and infrastructure. Consequently defining distribution and relative abundance by comprehensive ground survey is difficult and expensive. This highlights the need for simple, cheap, automated methodologies to predict the distribution of species in use, or having potential for use, in commercial enterprise. The technique applied here uses a Geographic Information System (GIS) to make predictions of probability of occurrence using an inductive modeling technique based on Bayes' theorem. The study area is in the Maningrida region, central Arnhem Land, in the Northern Territory, Australia. The species examined, Cycas arnhemica and Brachychiton diversifolius, are currently being 'wild harvested' in commercial trials, involving sale of decorative plants and use as carving wood, respectively. This study involved limited and relatively simple ground surveys requiring approximately 7 days of effort for each species. The overall model performance was evaluated using Cohen's kappa statistics. The predictive ability of the model for C. arnhemica was classified as moderate and for B. diversifolius as fair. The difference in model performance can be attributed to the pattern of distribution of these species. C. arnhemica tends to occur in a clumped distribution due to relatively short distance dispersal of its large seeds and vegetative growth from long-lived rhizomes, while B. diversifolius seeds are smaller and more widely dispersed across the landscape. The output from analysis predicts trends in species distribution that are consistent with independent on-site sampling for each species and therefore should prove useful in gauging the extent of resource availability. However, some caution needs to be applied as the models tend to over predict presence which is a function of distribution patterns and of other variables operating in the landscape such as fire histories which were not included in the model due to limited availability of data.