In waste textile recycling, the rapid and accurate determination of fiber type and content is a key part of the recovery program. In this paper, 598 waste cotton-polyester blended fabrics were used as the research object, and the raw near infrared spectra (NIRS) of the samples were tested by portable NIR spectrometer. In the 1 400~1 700 and 1 900~2 200 nm NIR regions, there was a clear difference between the spectra of 100% cotton and 100% polyester samples, and these spectral differences were reflected in the various color fibers. At the same time, the reason why the slant spectrum was produced might be surface effect of the fabric, the coloring method and the fine particles adhered to the fiber surface. Dark samples tended to drift their spectral baselines in the shortwave region. After the derivative pretreatment, the baseline drift was basically eliminated, and the oblique spectrum showed normal spectral characteristics. The quantitative analysis model of waste cotton-polyester blend fabric was established by partial least squares (PLS) method combined with 1st derivative, Savitzky-Golay (S-G) smoothing, mean centering and orthogonal signal correction (OSC) method. In order to verify the reliability of the model, root mean standard error of cross validation (RMSECV) was calculated and 346 external samples were selected to test the model. The RMSECV of the model was 0.002, and the relational coefficient of calibration ( RC ) was 0.998, and the relational coefficient of prediction ( RP ) was 0.997, and the standard error of prediction (SEP) was 1.121, and the prediction accuracy of the model was up to 97%. The error of NIR predictive value and gravimetric determination was within ±3%, and the consistency between the two is more than 90% , while the error was within ±5%, and the consistency was above 95%, and the analysis time of each sample was less than 10 seconds. Therefore, the waste cotton-polyester blend fabric fiber content could be quickly and accurately predicted by using NIR technology combined with the model.