Abstract: In order to determine the contents of palmitic,linolenic and linoleic acids in tobacco samples rapidly,the calibration models were established by relating the NIR spectra of 242 tobacco samples to their GC determined acid contents with partial least square method.The effects of the spectrum pre-treatment and the spectral range on the models were investigated,the internal cross validation and precision test,and external predictive validation with 30 other tobacco samples were conducted.The results indicated that:1) the multiplicative scatter calibration and the spectral range of 7502-5442.3 and 4427.9-4246.7 cm^-1,first derivative and the spectral range of 7502-5442.3 and 4601.5-4246.7 cm^-1,and vector normalization and the spectral range of 8724.7-5438.5 and 4427.9-4246.7 cm^-1 were suitable for palmitic,linolenic and linoleic acid models,respectively;2) the mean relative precision of the three models were 98.41%,97.84% and 98.84%,respectively;3) the predictive standard deviations of the established models were all less than 5%,and there was no significant difference between NIR prediction and GC measured values.The NIR models are suitable for rapid determination of palmitic,linolenic and linoleic acids in batches of tobacco samples.