Balance of correlations is an approach to build up quantitative structure-property/activity relationships (QSPR/QSAR). This approach is based on a split into the subtraining, calibration and test sets instead of classic split into training and test sets. The function of the calibration set is the preliminary check up of the model. In other words, the calibration set is like a preliminary test set. Computational experiments (with the Monte Carlo method) have shown that the statistical characteristics of the prediction for the toxicity to Tetrahymena pyriformis (the 50% growth inhibition concentration, IGC(50)) based on the balance of correlations are better than the statistical characteristics of the prediction based on the classic scheme.