Hyperspectral reflectances of flue-cured tobacco leaves were measured with ASD Field spec FR 2500 during different growth stages in different nitrogen, potassium treatments. Characteristic wavelengths were picked out through the correlations between nicotine and spectra reflectance, the first derivative spectra, spectral characteristic variables and the stepwise regression models. And the predicting effects of models were tested by the correlations between predicted values and mensurated values. The results showed that the first derivative spectra regression equation had the best predicting effect, and the spectra reflectance had better predicting effect, and the spectral characteristic variables had the worst predicting effect. The characteristic wavelength of spectra reflectance was 1792 nm and that of the first derivative spectra was 810 nm. The spectra reflectance and the first derivative spectra had an intimate relationship with nicotine content in 427～712, 741～810, 1382～1879, 1905～1969 and 2067～2338 nm. Accordingly, this study suggests that spectral data serve as promising measures to predict the nicotine content of tobacco leaves.