Using Fourier transform near-infrared (FTNIR) transmission spectroscopy and moving window partial least squares (MWPLS) method, the quantitative analysis models for brix degree of sugarcane clarified juice were established, and spectral analytical waveband was optimized. All samples were differently divided into the calibration set and the prediction set for 50 times, and the MWPLS models with changeable parameters were established for each division respectively, for the same parameters combination, then RMSEPs for all the 50 divisions were averaged. Based on the average data, the optimal model was selected, with the waveband, number of adopted wave numbers, PLS factor, RMSEP and RRMSEP were 7367-7324cm-1, 12, 8, 0.44oBx and 2.8%, respectively. The result shows the above prediction effect was obviously better than the one of PLS model on the whole spectroscopy collecting region, and the optimal model using only 12 wave number points was simple and stable, providing valuable reference for designing minitype special NIR spectral instruments.