To establish mathematical models for judging the aroma types of upper and middle flue-cured tobacco leaves, 128 samples (63 C3F and 65 B2F) from 11 main tobacco production provinces of China were selected as materials. Stepwise discriminant analysis was applied to samples with different aroma types and discriminant function was expressed with the proportions of 67 aroma components in total aroma constituents as the index. The results showed that the proportion of most aroma components in clear and full aroma tobacco leaves was higher than that in middle aroma leaves. The proportions of 51, 43 and 40 aroma components of clear, middle and full aroma tobacco leaves were higher in upper leaves than those in middle leaves. Aroma components dominated certain aroma types and differed between middle and upper leaves. The proportions of 18 and 11 aroma components in upper and middle leaves were led in the stepwise discriminant function respectively. Self-validation and cross-validation methods were applied to evaluate the original samples, and the accuracy rates reached 100% and 98.6% on middle leaves, 96.37% and 94.4% on upper leaves. The accuracy rates on some other samples reached 100% on middle leaves and 91.7% on upper leaves as predicted by the model. In conclusion, the proportion of aroma components as discriminant index could improve discriminant accuracy significantly in the middle and upper leaves. It could be used to analyse aroma types objectively, accurately and quickly.