为构建样品中的被测组分（TNT）的含量与其红外光谱之间的数学模型，从生产线上采集以及按相同方法制备了共计155个样品并采集了它们的红外光谱，根据计算所得光谱残留F值判别并剔除异常光谱。随机选取63个样品的光谱作为校正集，其余92个样品的光谱作为验证集。另外采用常规的溶剂提取-红外光谱法测定了这些样品中TNT的含量作为建模参考值。在最优模型波段（cm^-1）为：9 114.1~8 331.2，7 671.6~7 189.5，6 514.5~5 666，5 102.8~4 929.3，4 744.14~4 728.71的条件下，根据校正集的光谱数据，用偏最小二乘法建立数字模型。通过交叉检验均方根误差，RMSECV-维数曲线的理想程度以及光谱主成分分析结果选取了最优模型。采用χ²检验法，以及根据预测标准差和Bias值，结合验证集样品的光谱和数据，评估了方法的精确度和准确度。从TNT含量在36.68%~46.95%内的8个样品的测定结果得出其预测值的Bias值为0.078%，SEP%为0.514%。说明方法的准确度和精密度良好，且无需使用有机试剂。

To establish mathematical model correlating the analyte (TNT) in the explosive samples with their near infrared spectra, a total of 155 samples collected directly from the production line and self-prepared in the same way were taken, and their NIR spectra were taken with the abnormal spectra rejected according to the values of SpecResF found. 63 samples among the 155 were selected at random as the calibration set and the remainder (92 samples) were taken as the testing set. All these samples were analyzed by the conventional NIRS method with solvent extraction, to obtain values of TNT used as reference in modelling. Based on the NIRS data, PLS was applied to establish mathematical models, selecting the following optimum modelling wavelength sections:9 114.1-8 331.2 cm^-1, 7 671.6-7 189.5 cm^-1, 6 514.5-5 666 cm^-1, 5 102.8-4 929.3 cm^-1 and 4 744.14-4 728.71 cm^-1. Optimal models were selected through cross testing of RMSECV and considering the ideal of the curve of RMSECV/dimension and results of PCA of the spectra. Precision and accuracy of the proposed method were evaluated by applying the χ²-testing, and on the base of values of predictive standard deviation (SEP) and Bias values. 8 samples with TNT contents in the range from (w) 36.5% to 41.9% were analyzed by the proposed new method, giving value of Bias of 0.078% and SEP% of 0.514%. In addition, no organic solvent was used in the new method.