To study the relationship between chemical composition and flavour type of flue-cured tobacco, 500 tobacco samples were collected from 29 areas in 14 Chinese provinces in 2012. A classical sampling method was applied. Following relevant methods described in industrial standards and the published literature, we determined 117 chemical indicators which significantly affect the quality of flue-cured tobacco. The obtained data were processed by using multiple factor analysis for dimension reduction. A Bayes quantitative discriminant model was established and verified according to the scores of factors. A correspondence analysis between flavour type and planting area was also conducted. The results showed that 24 common factors, extracted from the original indicators, could explain 83.7% of the total variance. The quantitative discriminant model accurately predicted the flavour type of tobacco leaves on the basis of megastigmatrienone, furan, alkali and other substances. The accuracies of internal cross validation and external validation were ≥ 89.5% and ≥ 97.0%, respectively. The results of the correspondence analysis between flavour type - discriminated by the established model - and planting areas of the tobacco were generally in line with the results of sensory evaluation. Among the leaves from 14 provinces, those from Yunnan represented best the typical fresh flavour style, followed by those from Fujian and Sichuan; for leaves of medium flavour style, the planting areas were ranked as follows: Chongqing > Guizhou > Heilongjiang > Hubei; the robust flavour style was best in Henan, followed by Anhui, Guangdong, Jiangxi, Shaanxi, Shandong, Hunan. The discriminant model proved to be suitable for a fast and objective quality evaluation of tobacco leaves.