Abstract: The near-infrared(NIR) spectrums of pure cotton mixed with polyester, acrylic and silk fibers were measured in virtue of diffuse reflection technology with improved diffuse reflection sample carring accessory.Three prediction models were made respectively for three mixed samples with same commensurate cotton but different contents,and there were 7 samples for each prediction set which was used to verify prediction models.The results have proved that when the sample background of the calibration set was consistent with the prediction set,the predictive precision will be raised.Accordingly,the sample background was strongly influential to the model,and that could be eliminated by using correlation analysis.The root mean square error of prediction(RMSEP) was reduced from 4.267 7 to 2.896 5,and the prediction correlation coefficient was improved from 0.948 8 to 0.977 5,that proves the correlation analysis could enhance the prediction precision.