Near Infrared Spectroscopy has been widely used in material composition analysis.In this self-designed filter NIR device，the article presents a modeling method based on BP neural network with wavelength optimization selection.First,multiple linear regression algorithm modeling is used to analyze the prepared 26 samples to obtain optimal wavelength which is then used as BP neural network model′s input.After the calculation,it shows that the fitting residual is 8.768991×10-6 and the correlation coefficients of modeling samples and testing samples are respectively 0.994 and 0.996.The experimental result indicates that the BP neural network modeling method based on wavelength optimization can gain the optimal solution more quickly,reduce the variables used in modeling and apparently improve the robustness of the quantitative analysis model,enhance the actual ability of detecting.