Next, spectra normalised by the intensity of a selected band of CH3CN added as an internal standard to the studied samples were utilised. To compare the predictive ability of the models constructed, the relative standard error of prediction (RSEP) was calculated. The errors found for multivariate calibrations were a few times smaller than those for the univariate ones. Usually, the most effective was the PLS method, for which RSEP values of the order of 1–2% for calibration and 2–3% for testing data sets were obtained.