The knowledge of thermal behavior of combustion engines is extremely important e.g. to predict engine warm up or to calculate engine friction and finally to optimize fuel consumption. Typically, thermal engine behavior is modeled by using look-up tables or semi-physical models to calculate temperatures of structure, coolant and oil. Using look-up tables can cause in inaccurate calculation results because of interpolation and extrapolation and semi-physical modeling lead to high computation time. This study introduces a new kind of model to calculate thermal behavior of combustion engines by using artificial neural network which is high accurate and extremely fast in both building and calculating results. The applied neural network is a multilayered feedforward network and is trained by data which are generated with a validated semi-physical model to calculate the output data (e.g. oil temperature) by nonlinear transformation of input data (e.g. engine torque, engine speed or ambient temperature) and weighting of these transformations. It is described how the relevant input data have been defined and generated by performing a design of experiment with the semi-physical model. The training of the artificial neural network is described, too. The quality of the training is shown based on several statistic parameters (e.g. root-mean-square error). The validation is made based on measurement data. Therefore a comparison of calculated and measured temperatures for both steady-state and transient engine operating points is shown. Furthermore, the computation time is shown and compared to the physical model. Finally, the influence of variation of input data like engine torque, coolant volume flow or ambient temperature is investigated using the new model.