Numerous techniques exist for modeling landslide susceptibility including heuristic, statistical and deterministic analyses. More recently, knowledge-based techniques have been explored including data mining approaches whereby key data sets are assessed to establish inter-relationships with the primary training set, in our case, landslides. This paper analyses a study area of approximately 800 km2 on the Bellarine Peninsula in Victoria, Australia where landslides are restricted mainly to the coastal fringes and as such, form a ‘rare data set’ for the overall region. This paucity of training data presents problems for traditional susceptibility methods and, as a result, a series of trials using various data mining techniques were undertaken to assess their applicability to modeling susceptibility in the study area. A range of data mining techniques including Random Forests and decision trees implemented in the WEKA package (developed by the University of Waikato, New Zealand) as well as the See5 algorithm were applied to the data. Early results generated by these methods demonstrated the need for more sophisticated methods of pre-processing and selecting training data. Further discussion is also included on the various techniques used to analyze statistical accuracy of each method and their applicability to the prediction of landslide susceptibility through the production of susceptibility maps. Finally, the paper briefly discusses the challenges of a cross-discipline process where the highly statistically based mathematical approach of the analytical scientist must be combined with the skills of the geoscientist dealing with an uncertain real world situation where limitations in data availability and quality require a significant degree of expert judgment.