Abstract: The leaf structure and oil of 250 samples of flue-cured tobacco were valued with the appearance quality evaluation method, the samples were scanned in a wavelength range of 3500-12000 cm^-1 with near infrared(NIR) spectrometer, further, the NIR predicting models of structure and oil of tobacco leaves were established with partial least square method.The results showed that: 1) the vector normalization method and multiplication scattering correction method were suitable to setting up prediction models for leaf structure and oil, respectively;2) the models established through scanning within the wavelength ranges of 4246-4601cm^-1 and 6098-9222 cm^-1, 4246-5450 cm^-1 and 6098-7502 cm^-1 were better for leaf structure and leaf oil prediction respectively;3) the prediction accuracy was higher when the ranks of structure model and oil model were 13 and 16, respectively;4) the prediction accuracy of the models could be improved by rectifying the model parameters;5) the prediction accuracies of structure model and oil model were about 90% and 77%, respectively.The conclusion was that the structure model was applicable to leaf structure prediction, while the oil model needed to be further improved.