Thermal decomposition of automotive polyurethane was investigated by thermogravimetry under non-isothermal and isothermal conditions. For isothermal treatment, a neural network (ANN) was adopted with kinetic models as activation functions for neurons in the hidden layer. In this network architecture, rate constants represent weights between the input and intermediate layer and the learning process occurs by optimizing only the weights in output layer. Polyurethane sample was collected from an automotive intake manifold and the Diffusion and Contraction models were selected for better describe the decomposition as a combined event. Due to mathematical corrections, the accuracy of ANN is greater than individual model analysis. To validate the isothermal results, the non-isothermal analysis was performed and activation energy was calculated by Friedman, Flynn-Wall-Ozawa and Kissinger-Akahira-Sunose methods. The Ea calculated is 185-198 kJ mol(-1) for all the methods. The results were used in a critical analysis between the both methods.