Thermal decomposition process of automotive polyurethane was investigated by thermogravimetric analysis under non-isothermal and isothermal conditions. For the isothermal treatment, a methodology based on multilayer perceptron neural networks was adopted and kinetic models were used as activation functions for neurons in the hidden layer. [1] A polyurethane sample was collected from an automotive intake manifold and among fifteen models, the Diffusion and Contraction models were selected as a preferable set to describe the process. This choice was based on the network residual error in the experimental data description. The neural network architecture allows quantifying the contribution of each model in the entire process. [1] Because of the mathematical corrections in the kinetic functions, the accuracy of the network is greater than the individual model analysis. For the isothermal treatment, air atmosphere was considered and the experiments were performed at 380, 390, 400 and 410oC. In the non-isothermal analysis, the activation energy was calculated by Ozawa [2], Friedman [3], Flynn-Wall-Ozawa [2] and Kissinger-Akahira-Sunose [4] methods. For these experiments, air atmosphere was also used at four heating rates: 5, 10, 15 and 20 oC min-1. Thermodynamic properties of the decomposition reaction were determined and the activation energy was calculated as approximately 198 KJmol-1 for all the methods. The results were compared and analysed to predict thermal aging decomposition and to evaluate the polyurethane lifetime in vehicles. Moreover, a critical analysis between the isothermal and non-isothermal methods was performed to discuss their accuracy and performance.