In order to solve the problem of consumers misjudgment by hitting watermelon because of different watermelon varieties, a vibration detection system was built to get the frequency response functions of “Huang guan”, “Zaojia8424”, “Jng xin” and “Xisha” watermelon. Using the PNN neural network model achieved the identification of watermelon varieties, and the accuracy was more than 92%.At the same time,according to the relationship between the sugar content and the main resonant peak frequency, the prediction model of the sugar content of each variety of watermelon was established by stepwise multivariate linear regression analysis, and the coefficient were all above 0.86.The prediction set samples were identified and then predicted the sugar content. The measurement error of the sugar content of the watermelons which were identified accurately was less than 6.2%. The measurement error of the sugar content was larger for the watermelon which was identified mistakenly because of error prediction model. On the above basis, the unknown watermelons in the market could be identified, and then the appropriate corresponding model was selected automatically to measure the sugar content.