Internal combustion engines may be loaded at a test bench as in a vehicle to simulate driving cycles. The significance of these experiments is a direct consequence of the precision of the control system. Mathematical models of the test bench can on the one hand be used to simulate the system behavior and serve as starting point for model-based control on the other hand. Typically physical models are used for these purposes. In contrast, this paper presents the identification of a blackbox model of a dynamic internal combustion engine test bench directly from measurements. Several operating regions described by the accelerator pedal position and the engine speed were defined, local models were estimated and combined to a global model covering the entire operating region of the engine. No specific assumptions about the structure of the model have to be made, furthermore, the measuring time can significantly be reduced. A comparison with measurements on a test cycle covering a broad operating region show the good performance of the proposed model.