The artificial neural networks (ANN) are shown to be useful for quantitative X-ray fluorescence spectrometry of Ti, V, Fe, Ni and Cu in polymetallic ores. The performance of ANN was compared with the univariafe linear regression (ULR) model. The comparison was made on the calibration and prediction sets previously published. For study of performances of both calibration methods the mean squared errors and relationships between predicted and desired values were calculated. ANN were found to perform generaly better than the univariate linear regression model.