The quality evaluation of nature polysaccharides is a tough nut to crack because of its high Mw distributions and larger polarity property. It is well-known that infrared spectroscopy and multiple regression modeling have been used for quantitative examinations in multiple fields, but it has not been applied to the compositional analysis of polysaccharides. In this study, attenuated total reflectance-fourier transform infrared spectroscopy is used to simultaneously quantify aldoses, ketose and uronic acids in Atractylodes polysaccharides by a combination of multivariate regressions. After experience of different data processing pretreatments, the resulting spectrum contains maximum amount of information of monosaccharide contents in Atractylodes polysaccharides. In this case, different smoothing points, derivatives, SNV and MSC are used in the pre-modeling spectrum processing and VIP screening is used to reduce the number of variables to simplify the calculation of the model. All the most optimal prediction models have both good prediction ability (R2 ≥ 0.9 and RPD > 3) and no over fitting (RMSEP/RMSEC < 3). This strategy has opened a new possibility for the nondestructive determination of complex monosaccharide compositions of natural polysaccharides in a short detection time, low equipment requirement and high experimental safety.