In this study, a qualitative discrimination model was established based on the combined technology of near-infrared (NIR) and electronic nose (EN) to distinguish the light, intermediate and strong flavors of tobacco leaves. The results showed little difference in the accuracy of the three models, all of which were more than 89.00%. However, the prediction accuracy of the combined model for intermediate flavor and strong flavor was 82.67% and 80.00%, respectively, which were significantly higher than those by NIR (72.41% and 73.33%) and EN (68.97% and 53.33%). The reason may be that EN was more sensitive to aroma components affecting intermediate flavor and strong flavor, and captured more information. The new information as a beneficial supplement to NIR data and can be used to establish a model with higher accuracy for tobacco flavor classification. In addition, based on the same fusion data, this study compared the modeling and prediction accuracy of different data mining algorithms. The results showed that the modeling accuracy of the artificial neural network (99.07%) was higher than that of the support vector machine (96.26%). However, the prediction accuracy of the artificial neural network (65.00%) was significantly lower than that of the support vector machine (83.75%), which verified that the support vector machine could reduce overfitting in the modeling process. This study can support the rapid identification of tobacco flavor style, and the further development of this technology will strive to provide an auxiliary identification method for professional tobacco evaluators.