Soil hydraulic conductivity (K s ) is a crucial soil physical property that not only influences soil hydrological processes, but also the planning for vegetation recovery, irrigation practice and drainage design. However, K s data are often lacking at large-scale soil database due to difficulties in direct measurement that is often labour intensive, time consuming and cost inefficient. The objective of this study was to compare the performance of different emerging methods [Multiple linear regression (MLR) and artificial neural network (ANN)] of K s prediction. The pedotransfer function (PTF) is one such method that is based on selected factors closely correlated with K s at regional scale. We collected disturbed and undisturbed soil samples in the 0-40 cm soil layer at 243 sites across the entire typical Loess Plateau of China (430,000 km 2 ) and then measured K s and the potentially related factors. The results showed that K s was normally distributed with moderate a spatial variation (CV = 67%). Correlation analysis indicated that bulk density (BD), saturated soil water content (SSWC), clay content (Clay), silt content (Silt) and latitude were closely correlated (p b 0.05) with K s . Although the accuracies of MLR and ANN were equal in terms of estimating K s , the stability of PTF developed via ANN was not as good as that of MLR. Thus PTF developed via MLR, which included BD, Silt and Clay, was considered as the best model for estimating K s . There is a need to closely monitor the stability and repeatability of PTF during comparison and determination of PTF.