Various transport models are presently used to predict the long-term migration behaviour of fallout radiocesium on the soil. To examine to what extent the uncertainty of these predictions is influenced by the spatial variability of the migration rates, we determined the depth profiles of Chernobyl-derived 137Cs at 100 plots in a 100 m x 100 m pasture. These data were used to obtain the frequency distributions of the characteristic transport parameters of three widely used transport models (e.g. dispersion-convection model, residence time model, and back-flow model). The results show that these transport parameters are generally log-normally distributed with a coefficient of variation of about 80%. Finally, each transport model was employed to predict the resulting frequency distribution of the 137Cs inventory in the main root layer (0-7 cm) of the pasture, 20, 50, and 100 years after the deposition. If only the spatial variability of the transport parameters is taken into account, this analysis revealed that the dispersion-convection model and the back-flow model always predicted rather similar, but significantly higher median inventories than those obtained with the residence time model. If, in addition, the spatial variability of the amount of 137Cs deposited is also taken into account, the frequency distributions of the 137Cs inventories in the root layer become so wide that differences in the median inventories predicted by the three models become statistically significant only after 100 years. Several statistically significant correlations between the transport parameters of the three models were also detected.