Pearson correlation coefficient was used to choose some physicochemical indexes of grape which have strong correlation with those of wine and multi factor regression equations was established to determine their quantitative relations by the stepwise regression. Each physicochemical index of wine has a specific linear relationship with several physicochemical indexes of corresponding grape or just only one. At the same time, the multi-kernel support vector machine was carried out to classify the wine samples. The results from the multi-kernel support vector machine are approximately consistent with those from the artificial with an accuracy of 91.89%. Results from this study show that the physicochemical indexes of grape and wine can determine the taste evaluation of wine well.