In order to study the stability of chemical components in tobacco in processing, the near infrared (NIR) spectra of cut tobacco were online collected at the outlet of HXD drier and then processed by continuous wavelet transformation. The relationships between routine chemical components and NIR spectra of samples were studied on the basis of principal component analysis (PCA). The characterization models of tobacco with regard to the stability in processing were developed, then validated by changing the processing parameters of HXD drying. The results showed that the information of tobacco expressed by its routine chemical components could be characterized by NIR spectra. With a significance level of 0.05, the recognition rates to normal samples and abnormal samples for the characterization model of brand B were 98 % and 100 %, respectively. The recognition rates to normal samples for the models of brand A, C, D, E were 98.7 %, 99.3 %, 100 %, 98.4 %, respectively. The NIR model could effectively and real-timely monitor the quality of tobacco in processing.