Abstract: In order to promote the efficiency in discriminating flue-cured tobacco leaf categories, a new method based on thermal analysis spectra and machine learning was developed. The thermal analysis spectra from eight leaf flavor types were determined by a thermogravimetric analyzer. Via comparing the pyrolysis temperature differences among the eight flavor types, the characteristic pyrolysis temperatures for the different leaves were extracted. A flavor type discriminant model was constructed based on the support vector machine (SVM) improved by a genetic algorithm (GA), and its discriminant accuracy was verified. The results showed that:1) There were significant differences in the thermal analysis spectra of the tobacco samples in 150-400℃. 2) The pyrolysis characteristic temperatures for type Ⅰ-Ⅷ were 368.3, 763.4, 613.0, 517.2, 611.2, 652.6, 336.1, 383.5℃ respectively. 3) The flavor type determination model constructed with GA-SVM method had a prediction accuracy rate of 83.3%.