Atterberg limits and indices, for example liquid limit (LL), plastic limit (PL), linear shrinkage (LS) and plasticity index (PI), are important soil properties in engineering and land evaluation for predicting soil mechanical behaviour. This study was conducted to develop and evaluate pedotransfer functions (PTFs) to predict Atterberg limits using measured and morphological soil properties from a large data set in South Africa covering a vast range of soils, geologies and climates. Five PTFs were developed; the first four using measured properties from 2330 soil horizons including extractable Fe, Al, Mn, Na, K, Mg and Ca; organic carbon (OC); pH (H2O); cation exchange capacity (CEC); and sand, silt and clay fractions to predict LL, PL, LS and PI. Morphological descriptors such as colour, structure (grade, size and type), consistency, occurrence of slickensides and cutans and abundance of roots were included in the second PTF using data from 717 horizons to predict PI. For all PTFs, two‐thirds of the data were randomly selected and used for model development and the remainder for validation. Prediction accuracies of R2 between 0.49 and 0.77 comparable to other studies on large data sets but underperformed when compared to localized data sets. For engineering purposes, site‐specific PTFs for prediction of Atterberg limits should be developed.