A new approach of using visible-near infrared spectroscopy combined with different chemometric methods was investigated for the prediction of sugar content of rice wine. 240 wine samples were used for the calibration set, while 60 for the validation set. After some pretreatments of smoothing, standard normal variate (SNV) and first derivative to the spectral data, four different models were developed and the precision were compared, including partial least squares (PLS), combination of wavelet transform and PLS (WT-PLS), combination of principal component analysis and artificial neural network (PCA-ANN), and combination of PCA and least squares-support vector machine (PCA-LS-SVM). According to the standards of correlation coefficient (r), the root mean square error of prediction (RMSEP) and bias, a best calibration model of PCA-LS-SVM was achieved for the prediction of sugar content of rice wine. The correlation coefficient r=0.962, RMSEP=0.021 and Bias=-0.001 by PCA-LS-SVM, and a satisfying prediction precision was achieved. The results indicated that visible-near infrared spectroscopy could be successfully applied for the prediction of sugar content of rice wine and PCA-LS-SVM model could achieve a best prediction results.