Abstract: Based on thermal pyrolysis-gas chromatography-mass spectrometry (Py-GC-MS) and random forests (RF), the applicability of heat-not-burn tobacco raw materials was analyzed from the perspective of chemical composition. The sieved sample powder (0.90 mg) was placed into the sample cup, Py-GC-MS was used to detect 28 different types of heat-not-burn tobaccos, and MZmine software was used to process the Py-GC-MS data to obtain the characteristic peak table containing intensity information. The characteristic peak table and the sensory evaluation scores of the samples were used as independent variables and dependent variables, respectively, and RF regression algorithm was used to establish the applicability model of heat-not-burn tobacco raw materials. It was shown that the coefficient of determination of RF model on the training set was 0.93, and the root mean square error was 0.85. The coefficient of determination on the testing set was 0.92, and the root mean square error was 0.96. According to the qualitative results of NIST 2017 database, 20 chemical compositions with higher feature importance scores were screened out.