<p>Data on partition coefficients between sorbent materials and water are of great importance for the analysis of adsorption behavior and concentrations of dissolved organic compounds in the environment. The experimental definition of the partition coefficients is a complex enough process. Consequently, computational methods simulation of the values of the above partition coefficient is attractive alternative to direct experimentation. In this study, the polydimethylsiloxane-water partition coefficients of organic compounds (<i>n</i> = 284) were considered as endpoints in quantitative structure - property relationships (QSPRs). Using the CORAL software based on the Monte Carlo technique, different splits were built for different splits into the training and validation sets. The correlation intensity index (CII) is applied as an additional parameter of the optimization of correlation weights used in the calculation of the optimal descriptors. It is suggested that CII can be a criterion of the predictive potential for QSPR models. The statistical characteristics of the models for the validation sets are quite good: average determination coefficient on five random splits is as follows: 0.964 ± 0.004. The reliability of the approach was checked with several splits into the training and validation sets.</p>