In recent years, a new type of value called the Kansei value has been widely discussed, especially in the manufacturing industry. Many studies have tried to quantify Kansei values and found that human psychological and physiological reactions have certain effects. To further examine the relationships among psychological reactions, physiological reactions, and Kansei val-ues, the main purpose of this study is to construct a predictive model for beverage preference evaluation. However, we found that the problem of how to obtain psychological and physiological reaction data without affecting people's Kansei value evaluations has not been solved well. Therefore, this study proposes using an arm robot instead of a human to perform a series of actions when drinking, which solves the problem of measuring psychological and physiological reactions simultaneously. Using the psychological and physiological reactions collected by this method, we completed a predictive model of people's preference evaluations. Our model's prediction accuracy is higher than that of the model using psychological evaluation or physiological reaction alone. This result proves again that psychological and physiological reactions do affect people's preference evaluation simultaneously. Through the model comparison, we also get the result that the evaluation of people's preference for beverages has been formed in tasting the beverages in the mouth.