This paper presents a new methodology for the detection of critical situations for vehicle lateral dynamics, in order to improve the ESP systems. The principle of the approach rests on the modeling of the vehicle thanks to a bicycle model. The lateral tire forces are well known to be highly nonlinear. However, in normal driving region, those force can be approximated as linear. The detection of the critical situation is done when the physical data provided by the sensors on the vehicle do not anymore match with the linear tire force. The problem of the detection of critical situation is so rewritten as a problem of a fault detection, where some dedicated approaches are given in the literature. Since the vehicle model depends on its longitudinal speed, an extension of the parity space approach for linear parameter varying system is handled. The results has been applied on a real vehicle within the INOVE project * in order to detect a loss of stability in the vehicle.