Abstract: Taking 90 flue-cured tobacco samples of 3 flavor styles as the research objects, the flavor styles of the samples were determined by sensory evaluation. The samples were extracted by dichloromethane, and the characteristics of GC/MS fingerprints were analyzed by VIP (Variable importance in the projection) method combined with PLS-DA (Partial least squares discriminant analysis). The aroma components influencing the flavor style of flue-cured tobacco were analyzed, and the classification models of the fingerprints and aroma components were built separately. The correlations between the characteristic aroma components and main aroma note of flue-cured tobacco were analyzed, and the aroma components influencing the main aroma note were investigated. The results showed that:1) 1 533 pieces of information of characteristic fingerprints were screened out, and the validation accuracy, R², RMSECV and RMSEP of the developed GC/MS fingerprint discrimination model were 96.67%, 0.926, 0.190 and 0.250, respectively. 2) 52 aroma components greatly influencing the classification of flue-cured tobacco flavor style were identified, and the validation accuracy, R², RMSECV and RMSEP of the developed aroma component classification model were 86.67%, 0.817, 0.301 and 0.396, respectively. 3) 16 aroma components were introduced via stepwise regression analysis, and the discrimination equations for flue-cured tobacco flavor styles (robust, fresh and medium flavor styles) were established. The back substitution verification indicated that the discrimination accuracies of these equations were 100%, 91.7% and 97.2%, respectively. 4) The selected 16 aroma components significantly correlated to the main aroma note of flue-cured tobacco. All above mentioned discriminant models could identify the flavor style of flue-cured tobacco, the accuracies of characteristic fingerprint model and stepwise regression discriminant equations were higher than that of classification model based on 52 aroma components.