In the initial phase of a national project to map clay, sand and soil organic matter (SOM) content in arable topsoil in Sweden, a study area in south-west Sweden comprising about 100 000 ha of arable land was assessed. Models were created for texture, SOM and two estimated variables for lime requirement determination (target pH and buffering capacity), using a data mining method (multivariate adaptive regression splines). Two existing reference soil datasets were used: a grid dataset and a dataset created for individual farms. The predictor data were of three types: airborne gamma-ray spectrometry data, digital elevation from airborne laser scanning, and legacy data on Quaternary geology. Validations were designed to suit applicability assessments of prediction maps for precision agriculture. The predictor data proved applicable for regional mapping of topsoil texture at 50 9 50 m 2 spatial resolution (root mean square error: clay = 6.5 %; sand = 13.2 %). A novel modelling strategy, 'Farm Interactive', in which soil analysis data for individual farms were added to the regional data, and given extra weight, improved the map locally. SOM models were less satisfactory. Variable-rate application files for liming created from derived digital soil maps and locally interpolated soil data were compared with 'ground truth' maps created by proximal sensors on one test farm. The Farm Interactive methodology generated the best predictions and was deemed suitable for adaptation of regional digital soil maps for precision agricultural purposes.