[en] Scatter of data found in a sorption database may bring about possible risk dilution in the result drawn by the probabilistic safety analysis unless inappropriate data were carefully examined and excluded if necessary. To reduce such scatter, the advantage of data analysis by adopting theoretical sorption models was demonstrated. The sorption coefficients (K_d) of various combinations of bentonite, minerals and rocks with elements in the database were examined at first to extract the data sets that have enough information for the analysis using the mechanistic sorption models such as the ion exchange and the surface complexation. The K_d of Cs on bentonite was selected as an example and the factors dominating the data scatter were found to be the concentrations of Cs and coexisting ions. The procedure to estimate the parameters that were not given in the database, such as the selectivities of ion exchange reactions among Cs and coexisting ions, and the cation exchange capacity were also suggested to assist in the model analysis. Although the K_d of Cs varied in several orders of magnitude, the scatter could be reduced within the same order of magnitude after appropriate interpretation of the reason for the scatter and data selection. (author)