In this paper, an ATV (All-Terrain Vehicle) rollover prevention system is proposed. Dynamic instability evaluation is based on the on-line estimation and prediction of the Lateral Load Transfer (LLT) from a vehicle model based on two 2D representations. As off-road vehicles are considered, grip conditions have a large influence. They are here estimated relying on observation theory. Nevertheless, two main behaviours (over/under-steering) may be encountered pending on grip and vehicle configuration. Since only a low cost perception system can be considered in ATV applications, these two opposite dynamics cannot be explicitly discriminated. As a result, two observers are designed, according to the vehicle behaviour, to estimate on-line the terrain properties (grip conditions, global sideslip angle and bank angle) and a “supervisor” selects on-line the right observer. Next, a predictive control algorithm, based on the extrapolation of rider's action and the selected estimated dynamical state, allows the rollover risk to be anticipated, enabling to warn the pilot and to consider the implementation of active actions. Simulations and full-scale experimentations are presented to discuss the efficiency of the proposed solution.