In order to improve the efficiency of quality evaluation during the bidding process of tobacco flavors, quality discrimination models for 395 tobacco flavors were established by scanning flavors with their near-infrared spectra and backed by a consistency evaluation method. The models were applied to evaluate the quality consistency of the bidding flavors. The results showed that: 1) The selected wavelength range was from 4 000 cm-1 to 9 000 cm-1, and the first derivative was combined with the Standard Normal Variate (SNV) to pretreat the spectra with a CI (Conformity Index) threshold of 3. 2) The threshold of near-infrared spectral similarity matching degree (M) was set at 90%, so that the probability of misidentification was extremely low. 3) Using the models to evaluate the quality of the bidding flavors, 92.2% of the bidding samples (364 tobacco flavors) were discriminated quickly and accurately. The models could characterize the similarity between the bidding samples and the standard samples, which was consistent with the results of chromatographic fingerprint analysis. The method is fast, accurate, practical, and suitable for quick evaluation of the quality of flavors and fragrances in bidding.