A rapid detection model for 2,3-butanedione and 3-hydroxy-2-butanone in base liquor was established using near infrared spectroscopy(NIR).182 base liquor samples from Dukang winery in Luoyang were selected as materials.The chemical values of the two ketones were detected by gas chromatography(GC).Meanwhile,the spectral data in the range of 12 000-4 000 cm-1 were collected.The calibration models for 2,3-butanedione and 3-hydroxy-2-butanone were established by partial least square(PLS) combined with internal cross validation.By comparing effects of the PLS models under different spectral pretreatments for optimization,the optimal preprocessing methods for 2,3-butanedione and 3-hydroxy-2-butanone were determined as first derivative with multiple scattering correction and second derivative,respectively,while the optimal spectral ranges are 9 403.2-7 497.9 cm-1 and 9 403.2-7 497.9 cm-1 + 6 101.7-5 449.8 cm-1.The determination coefficients(R2) for the chemical values and the NIR predicted values of the 2,3-butanedione and 3-hydroxy-2-butanone calibration set samples after optimization were 0.960 2 and 0.963 2,while the corresponding root mean square errors of cross validation(RMSECV) were 0.39 mg/100 mL and 0.22 mg/100 mL,respectively.Through external validation,the R2 for the validation set samples were 0.957 6 and 0.957 8,while the predicted root mean square errors(RMSEP) were 0.40 mg/100 mL and 0.24 mg/100 mL,respectively.Results showed that the model established by near infrared spectroscopy combined with chemometrics could meet the requirements for rapid detection of ketones in liquor production with high accuracy.