Matrix effect always affects EDXRF (Energy Dispersive X-ray Fluorescence) analysis of mineral samples. In this work

characteristic X-ray peaks of Ti and Fe in different ore samples were fitted by Gaussian fitting function and analyzed in an attempt to solve nonlinear relation between the counting rate and elemental contents

Classification and forecast methods were used in the analysis. Core samples (80 groups) were classified by the SOFM self-adapt neural net and the identification rate of concentrated iron ore and concentrated titanium ore samples is 100%. The RBF net was used to forecast the contents of Ti and Fe. Comparing the forecast results with chemical analysis results

relative errors of 65.4% samples were less than ±1%

and the others were less than ±3%

whereas less than 5% is demanded by the industry production. The results show that it is an effective method to use ANN technique to classify and forecast element content in core samples.