In view of the problems of weak objectivity and low design efficiency for traditional tobacco leaf formulation design, a method of tobacco formulation design was proposed based on the quality characterization by near infrared spectroscopy. The prediction model including conventional chemical compositions, position and aroma type was established by partial least square method, to predict the quality indexes of tobacco leaves. Using quality index as multidimensional vectors a comprehensive quality representation index was constructed. According to the rule of “similar substitution”, a set of candidate tobacco leaves was formed refer to the original tobacco grade in the target module to screen similar substitute tobacco leaves. The proportion of tobacco leaf in the substitute module is calculated by linear programming method. Using a combination of indicator evaluation and sensory verification to the effectiveness was verified for Sichuan tobacco leaf formulation designed in 2020. The results showed that the relative error of the tobacco leaf position, total sugar, reducing sugar, total alkaloids, and total nitrogen content was less than 6% between the substitute formula and the target formula. Except for stimulating indicators, there was no significant difference in sensory quality indicators such as aroma content, clarity, diffusivity, group nature, impurities, aftertaste, and strength between the substitute formulation and the target formulation.