Objective: Accuracy of NIR model is restricted by complex matrix and overlapping bands of traditional Chinese medicine (TCM). To solve the above problem

the application of competitive adaptive reweighted sampling (CARS) in quantitative analysis of Chinese medicinal materials

TCM extract and patent medicines was discussed in this paper. Method: Near-infrared diffuse reflectance spectra of Puerariae Lobatae Radix

Puerarin Lobata Radix extract and Yufeng Ningxin dripping pills were collected to determine the contents of puerarin. Then the spectra were pre-processed with optimized methods to exclude uncommon samples

and CARS method was used to screen relevant variables and establish partial least squares (PLS) correction models. Result: Root mean square error of cross-validation (RMSECV) was 0.35%

1.76% and 0.54%

respectively in models for raw material

semi-product and finished product. As compared with the models based on the full spectra

increased accuracy was obtained from the CARS-PLS models. Conclusion: CARS could increase prediction accuracy of the models and simplify calibration

which would offer a new approach for rapid and non-destructive determination of TCM.