In this article we compare the performance of four nonlinear state observers for a combustion engine test bench simulator including combustion oscillations, noisy measurements and disturbed inputs. These observers are the high-gain observer (HGO), the sliding-mode observer (SMO), the nonlinear extended state observer (NESO) and the extended Kalman Filter (EKF). The different observers are compared in open-loop in terms of the mean quadratic estimation error, computing time and convergence rate. A first important result obtained is that the NESO performance is good, although it does not need to know the engine friction model. Then the best of these is compared in closed-loop with a partial Luenberger observer which requires the knowledge of the model of the combustion oscillations. It turns out that we can achieve similar tracking results without the knowledge of the combustion oscillations model. © 2009 AACC.