Abstract: To construct accurate and robust discrimination models for flue-cured tobacco from eight major ecological areas, flue-cured tobacco samples were collected from eight domestic ecological areas and 70 constants and semi-trace components in tobacco leaves were analyzed. The modeling indexes were screened using a stepwise selection method. Discrimination models for flue-cured tobacco from the eight ecological areas were constructed based on Fisher Discriminant Analysis (FDA) and contents of tobacco chemical components. The constructed models were used to discriminate aged tobacco strips in warehouses of sampled enterprises. The results showed that: 1) By using the stepwise selection method to screen the indexes and chemical components with a frequency of 70 as the modeling indexes, the average discrimination accuracy of the models was the highest, achieving 91.5% in the 5-fold cross-validation. 2) The discrimination models for flue-cured tobacco from the eight ecological areas constructed using the FDA method and chemical component contents had a discrimination accuracy of 94.3% in the training set and a discrimination accuracy of 93.0% in the test set. The proximity of the spatial projection positions of each ecological area was generally consistent with the proximity of the actual geographic locations and sensory styles of the leaf samples. 3) With the data of the 70 tobacco chemical component contents predicted by near-infrared spectroscopy, the above-mentioned modeling methods could be applied to accurately and robustly distinguish the actual aged tobacco strips of domestic and foreign flue-cured tobacco leaves in the warehouses of two enterprises.