Atterberg consistency limits (liquid limit, LL; plastic limit, PL; shrinkage limit, SL) and indices (plasticity index, PI; friability index, FI) are useful indicators of soil mechanical behaviour. This study was conducted to evaluate the use of soil and environmental data for predicting Atterberg limits or indices using artificial neural network (ANN) models at the watershed scale in western Iran. The LL, PL, SL, PI, FI, particle size distribution, organic matter (OM) and calcium carbonate equivalent (CCE) were measured in soil samples collected from 113 locations. Three sets of readily available properties were employed as inputs. The first of these data sets or models consisted of soil properties. The second included topographic attributes and normalized difference vegetation index (NDVI), and the third was a combination of soil, topographic attributes and NDVI. Developed ANN models could explain a majority of the variability (62–94%) in Atterberg limits and indices. Greatest and poorest performances were attributed to the third and first models, respectively. No significant efficacy difference was observed between the second and third models. Therefore, the second data set with its readily available environmental variables is suggested for use in predicting Atterberg limits and indices at the regional scale. Sensitivity analysis showed that NDVI, OM, clay content, CCE and topographic attributes (wetness index, elevation, plan curvature and slope) could explain much of the variance associated with Atterberg limits and indices at the watershed scale in western Iran.