Glass transition temperatures of polymers were modelled by means of the CORAL software available on the Internet (http://www.insilico.eu/coral). The architecture of monomers was represented via simplified molecular input line entry systems (SMILES). Three random splits into the training and validation sets were tested to build up quantitative structure - property relationships (QSPRs). The index of Ideality of Correlation (IIC) represents a new measure of predictive potential. Application of the IIC as the criterion of predictive potential for the calibration set was tested in this work and resulted in correct recommendations for selection of the best model from three different models considered here.