Codonopsis root, a traditional Chinese medicine, produces saccharides that serve as key indicators for assessing its quality. In this study, we employ near-infrared reflectance spectroscopy to predict total polysaccharide concentration as well as the major individual saccharides of intact (surface and cross-section) and ground samples of Codonopsis roots. We developed regression relationships between the spectra and the concentration of fructose, glucose and polysaccharides for the ground spectra set (Rp2 > 0.8, RPD>2.0) based on partial least squares regression (PLSR) model. However, for the intact cross-section spectra set, only the dominant component of the fructose, could be quantitatively detected (Rp2 > 0.8, RPD>2.0). All models based on the intact surface spectra set were of unacceptable performance. This discrepancy could be attributed to the low concentration levels of different saccharides and the distinct chemical profiles reflected by the spectra sets from samples with varying homogeneity. The findings provide an approach for efficient and sustainable analysis for the purposes of quality control and standardization of Codonopsis roots.