Thalidomide is a teratogenic substance that was developed in 1950 for use as a sedative. However, only in 1961 it was associated with poor fetal formation due to maternal use [1]. The kinetic study of its thermal decomposition will be carried out by two methodologies: (a) adjustment of physical models to experimental data of decomposition fraction as a function of time, obtained in isothermal conditions and (b) adjustments of functions that correlate the heating rate and the temperature in which certain conversion fractions occur. For this analysis, experiments are required under different heating rates. However, it can be observed that, in some cases, the adjustment obtained by individual models under the whole experimental decomposition curve can present quite high error values and the adjustment methods in non-isothermal experiments do not allow defining the physical mechanism. For a more complete analysis of the whole process, this work proposes the use of a multi-layer neural network for the adjustment of the isothermal experimental data. This network was developed to determine the contribution of each model in the total fit and best fit region of each model. For the non-isothermal conditions, also called isoconversional methods, the activation energy was calculated from the models Ozawa [2], Friedman [3], Flynn-Wall-Ozawa [2] and Kissinger-Akahira- Sunose [4]. For the isothermal conditions, the experiments were carried out at temperatures close to the temperature of the first mass loss event of the material. For the nonisothermal treatment, the analyzes were done in inert atmosphere of nitrogen, with the heating rates of 2.5, 5, 10 and 20 ° C / s. The results obtained under the two conditions will be compared and will be used to predict the time of decomposition at room temperature, which allow the estimation of its shelf-life and metabolic activity of the drug in the body. In addition, the accuracy of the isothermal and non-isothermal methods will be discussed.