In this work two model based predictive controllers have been designed in order to regulate a non-linear liquid level system. First, a controller according to the Extended Prediction Self-Adaptive Control (EPSAC) was designed. This algorithm requires a linear model, so the model was linearized around a certain equilibrium point. This gives bad results when the setpoint lies far from the equilibrium output level. Secondly, a Non-linear Extended Prediction Self-Adaptive Control (NEPSAC) was designed. A big advantage is that no linearization is required. Consequently, a correct model is available at each point. This explains why NEPSAC gives the best results: a low settling time, no overshoot and equally good results for all setpoints. Finally, the performance of the controllers is evaluated, in order to carry out a tracking to a reference level and an effective rejection of the disturbances.