Regularities and peculiarities of gas chromatographic analysis of thermally unstable compounds were considered on the example of mixture of the reaction products of isopropylbenzene (cumene) free-radical chlorination. The principal constituent in this mixture is (1-chloro-1-methylethyl)benzene, which has the lowest thermal stability, and is partially converted to α-methylstyrene, the only product of its thermal destruction at the chromatograph injector temperatures up to 300 °С. Nevertheless, the results of the study confirms that the gas chromatographic analysis of chloroalkylarenes is possible without their decomposition with the injector temperatures up to 200 °С, even if the analytes contain chlorine atoms at the tertiary carbon atoms and in the “benzylic” positions relative to the aromatic fragment. Similar control of thermal stability of analytes can be recommended for other samples contained potentially unstable constituents. It is shown that thermal decomposition of thermally unstable constituents of samples cannot be revealed from the results of gas chromatographic analysis with capillary columns using variations of their absolute peak areas. Such task can be solved only by using relative peak areas calculated in respect to thermally stable compounds. The dependencies of relative peak areas of unstable constituents vs. temperature (descending), as well as those of their decomposition products (ascending) are characterized by presence of two limits. Low temperature limits correspond to the real content of unstable constituents or their decomposition products is the samples, while the upper limits – to the composition of such samples at their hypothetically complete destruction. Such dependencies can be approximated by logistic regression equation if sampling into capillary columns is carried out at relatively high split ratios (approx. not less than 10 : 1). At lower split ratios the temperature dependencies of peak areas of unstable constituents and products of their transformation are strongly distorted by so-called sample’s composition discrimination effects that make impossible data approximation using logistic regression.