Digital soil class mapping (DSCM) provides a means of meeting the growing global demand for soil information. The search for optimal models for digital soil class mapping to take advantage of increasing availability of ancillary information, such as gamma radiometric data, is ongoing. One of the novel approaches to DSCM is based on genetic algorithms, which provide predictive function for DSCM. This paper aims: to develop a scheme for implementing genetic algorithms for rule-set production (GARP) in digital soil class mapping; to compare the performance of GARP and classification tree model (CT); and to evaluate the usefulness of gamma radiometrics as a predictor for DSCM of legacy soil data. We first collated the legacy soil class data from databases of soil profiles and the associated ancillary data from disparate sources. We then created a 200-m resolution DSCM based on the Australian Soil Classification, for the Namoi catchment in north-western New South Wales, using GARP based on the general scorpan-sspfe model and compared the GARP performance with the widely used CT model. Elevation, terrain attributes, magnetic survey, land use, NDVI, and, where available, radiometric data were used as the ancillary variables. In this implementation, inclusion of radiometric data in either of the prediction models significantly improved the classification accuracy and the resulting DSCM. Based on various classification and prediction performance measures, GARP was shown to be outperformed by the CT. We conclude that GARP needs further improvement for its full potential to be realised for digitally mapping soil classes.