Abstract: The study was based on the aroma components of tobacco leaves to establish the classification model of tobacco flavor, and then all of the models were compared to select the optimal model. Firstly, detected 45 components tobacco leaves by tobacco industry standards, then selected 14 aroma components by stepwise regression method, discriminate analysis, Logistic regression, Gauss mixture model, classification tree, using K nearest neighbor method, artificial neural network and support vector machine seven methods to establish the models based on the 14 index. Using 100 randomly selected samples as the training sets and test samples to calculate the error classification rate through the establishment of the different methods of models, the model was the preferred model which classification error rate was lower than others. By contrast, two kinds of flavor function model (linear discriminate method and Gauss mixed) could be better to unknown sample types. Two kinds of optimization models had a certain application value for classification research of tobacco flavor.