Abstract：Outlier samples often strongly influence the stability of the model in near-infrared（NIR） spectroscopy. By using the method of PCA-mahalanobis distance, 12 outliers were detected and eliminated from 60 vinegar samples in calibration set. The calibration model with the remained 48 samples was established and used to estimate the contents of total-acid and volatile acid of the other 15 vinegar samples in predicted set. The results showed that the relative coefficients（R） of the predicted set were 0. 975 9 and 0. 938 3 ; the root mean square errors of prediction were 0. 093 8 and 0. 613 5 ; and the mean relative errors of prediction were 1.34% and 2.80%. Compared to the original calibration model,the predicted accuracy and stability of this model were greatly impoved.
2008, Vol. 29 
