Abstract: In order to measure total acid(TA) and total ester(TE) accurately and quickly in liquor, calibration models were established based on Fourier-transform near-infrared spectroscopy with artifical neural network by Cross-Validation after the spectra were analyzed and discussed. Results show that according to the selected spectral ranges of 6102～5446 cm^-1, coefficients of the models(R²) for TA and TE are 96.73% and 99.58% respectively, and the root mean square errors of cross validation(RMSECV) for TA and TE are 0.048 g/L and 0.085 g/L respectively. Then the models were tested. Results show that R² for TA and TE are 99.2% and 99.7% respectively, and the root mean square errors of prediction(RMSEP) are 0.074 g/L, 0.134 g/L respectively. These suggest the models of TA and TE are reliable with good predictability and can meet the requirement of quick determination of TA and TE in liquor production.