Abstract: Reconstituted tobacco (RT) has the advantages of low density, high filling value and low tar release, which can significantly reduce production consumption and control the release of tar and harmful components in cigarette smoke. However, some tobacco components will be lost during the processing of RT, while some non-tobacco components (such as pulp, etc.) will be introduced, which makes the sensory quality of RT different from that of natural tobacco (NT) cigarettes, that is, RT will produce an irritating odor when smoking, which directly affects the proportion of their doping in the preparation of cigarettes. In order to explore the key chemical components of irritating odor in the smoke of RT, elemental analysis (EA) and inductively coupled plasma mass spectrometry (ICP-MS) were used to characterize NT, RT, and substrates to analyze the chemical composition differences of the three raw materials. Then, an U-shaped fixed-bed reactor (UFBR) was used to conduct pyrolysis experiments on the three raw materials. The volatile products generated by pyrolysis were detected in real-time by a single photoionization mass spectrometry (SPI-MS), which was connected to the outlet of UFBR. Besides, the condensed liquid products were also characterized by gas chromatography-mass spectrometry (GC-MS) to provide the supplementary informations on chemical structures of detected compounds. In addition, the principal component analysis (PCA) was used to obtain the variance in SPI mass spectra and to distinguish the characteristic pyrolytic products of the three raw materials. Finally, the products with significant differences in concentration in the raw materials were identified as model compounds for sensory evaluation. The results showed that the key chemical components of irritating odor in RT smoke mainly contain cyclopentanone, cyclohexanone and 4-ethylphenol. Furthermore, the temperature range for the release of various model compounds during the pyrolysis process were determined based on the temperature-evolved profiles. This work provides a foundamental data and clues for future research on reducing the generation of irritating odors in RT.