Because of different constraints (such as different kinds of measurable elements, characteristic X-ray energy, changes in matrix composition, etc.), usually it’s not easy to get accurate information of elements, resulting in mistakes in later data analysis of energy disperse X-ray fluorescence measurement. The method is based on McCulloch-Pitts neural network (M-P neural network), according to matrix effect, to establish a new neural network model for quantitative forecasting of Zn by taking the data of X-ray fluorescence measurements of Cu, Fe, Pb, etc in lead-zinc mine in western Tianshan as the training sample. The relative error between predicted value and measured value is less than 5%. This method can be more accurate and rapid for X-ray fluorescence; it provides a new approach to correcting information of X-ray fluorescence.