Based on the principal component analysis, we established F score statistics to reflect comprehensive information about main chemical composition and their ratio and used F score as the characteristic value to characterize the homogeneity of formula tobacco. We detected the near-infrared spectral of tobacco samples with different formula tobacco and founded the near-infrared spectral model of F score by using the partial least square method based on the range of 4 281.2~4 485.6，5 542.4~6 024.5，6 043.8~7 135.3 cm^-1 spectroscopy pretreated by multiplicative signal correction and first-order derivative and Savitzkyt-Golay smoothing. The determinant coefficient of F score model was 0.996, and RMSECV was 0.269. The model was found to have good prediction performance and high prediction accuracy. The F statistics score of tobacco samples could be predicted quickly by NIR, and this provided a rapid and efficient monitoring method of tobacco blending uniformity and a technical guidance to the regulation and control of cigarette homogenization production.