Appraisal of quality is a crucial link in wine industry. At present, appraisal of wine quality mainly relies on sensory evaluation method, which may derive subjective biases. It shows particular importance for an establishment of objective and reasonable wine evaluation system to regulate the wine market. In this paper, after identifying the existence of multicollinearity of the physicochemical indicators of grape and liquor, Lasso regression is employed to build an evaluation model. Corresponding to the physicochemical indicators, this model is set up from four aspects: appearance, taste, aroma and overall balance. Lasso regression is used in each aspect. Weighted sum of the four aspects, overall assessment of the wine is obtained. As a conclusion, the new wine evaluation system is more objective and credible, thus it can substitute for sensory evaluation methods.