目的:采用近红外光纤光谱技术对复方益肝灵片进行快速判断厂家归属以及快速检定其中水飞蓟素的含量。方法:以全国6个企业生产的93批复方益肝灵片作为分析对象,建立聚类分析模型。光谱预处理方法为二阶导数和矢量归一化;平滑点:13;光谱范围:9 000~4 100 cm-1;光谱之间距离采用欧氏距离,光谱和类之间距离采用方差平方和法(Ward’s算法)。同样以全国6个企业生产的93批复方益肝灵片作为分析对象,建立定量分析模型。光谱预处理方法为一阶导数加矢量归一化;平滑点:17;光谱范围:9 100~7 300 cm-1和7 100~4 100 cm-1;采用偏最小二乘法回归。结果:聚类模型可以很好的区分训练集中的药品,在对作为测试集的30张光谱进行验证中全部判断正确。所建立的定量模型交叉验证均方根误差(RMSECV)为0.082 8,决定系数为97.98%,外部验证均方根误差(RMSEP)为0.044 1,R2为98.95%。结论:该聚类方法和定量模型快速简便、准确可靠,可以满足药品现场快速检查的需要。 OBJECTIVE: To rapidly determine the ownership of the compound Yiganling tablet and determine the content of silymarin in it by near-infrared spectroscopy. Methods: 93 batches of Yiganling tablets produced by 6 enterprises in China were selected as the analysis object, and a cluster analysis model was established. The spectral preprocessing method is second derivative and vector normalization; smooth point: 13; spectral range: 9 000 ~ 4 100 cm-1; Euclidean distance is used for the distance between the spectra; square sum of squares is used for the distance between the spectrum and the class; Method (Ward’s algorithm). Similarly, 93 batches of Yiganling tablets produced by 6 enterprises across the country were taken as the analysis objects to establish a quantitative analysis model. The spectral preprocessing method was normalized by the first derivative plus vector; the smoothing point was 17; the spectral range was 9 100-7 300 cm-1 and 7 100-4 100 cm-1; partial least-squares regression was used. Results: The clustering model can well distinguish the drugs in the training set, and all the judgments in the validation of the 30 spectra as the test set are correct. The RMSECV of the quantitative model was 0.082 8, the coefficient of determination was 97.98%, the root mean square error of external validation (RMSEP) was 0.044 1, and the R2 was 98.95%. Conclusion: The clustering method and quantitative model are fast, simple, accurate and reliable, which can meet the needs of rapid inspection on the drug site.