The quantitative models for alcohol and acid of Chinese rice wine were established by Near-infrared (NIR) transmission spectroscopy combined with Successive Projection Algorithm (SPA). Considering the influence of different kinds of wine, the optimal modeling method of alcohol and acid was developed by comparing the classification and mixed models for dry and semi-dry wine. SPA-PLS model was established by optimizing wavelengths on the range of 800~2500nm, which was better than full-spectrum PLS model. The correlation coefficient (R) of the model for acid,alcohol in dry wine and alcohol in semi-dry wine were 0.943, 0.983 and 0.993, respectively, and the Root Mean Square Errors of Cross Validation (RMSECV) were 0.258, 0.174 and 0.189, accordingly. The result indicated that the classification modeling significantly increased the precision of alcohol model, and the acid model has been improved by mixed modeling. SPA made the analysis model simple, and promoted the precision and stability. The result provides aneffective way of building robust quantitative model in Chinese rice wine, and can be applied in on-line wines’ quality testing.