The prediction of these contents plays an important role in the quality control of cigarette manufacturing and the development of new products. A novel method for simultaneous prediction of nicotine, tar and carbon monoxide (CO) in the mainstream smoke was developed based on the Fourier transform near infrared (FT-NIR) diffusing reflectance spectra of tobacco powder. The calibration models for nicotine, tar and CO contents were established by the help of partial least square (PLS) and these values measured by the traditional methods. The coefficients of determination of the obtained models are 0.9678, 0.9115, 0.9312, the root mean square errors of cross validation (RMSECV) are 0.0804, 0.6142, 0.2190, and the predictive ranges are 0.4255~3.000mg, 9.770~21.72mg and 10.67~15.37mg respectively. The method was capable of predicting the contents of nicotine, tar and CO in the mainstream smoke of various flue-cured tobacco samples with a mean relative deviation smaller than 5% and good reproducibility. The advantages of this method are that the sample preparation is very simple and the measurement of NIR is pretty fast. Moreover, this method is safe, economic and environmentally friendly, as it avoids the use of any chemical solvent.