In embedded systems such as electric vehicles, Proton exchange membrane fuel cell (PEMFC) has been an attractive technology for many years especially in automotive applications. This paper deals with PEMFC operation monitoring which is a current target for improvement for attaining extended durability. In this paper, supervision of the PEMFC is done using knowledge-based models. Without extra sensors, it enables a clear insight of state variables of the gases in the membrane electrode assembly (MEA) which gives the PEMFC controller the ability to prevent abnormal operating conditions and associated irreversible degradations. First, a new state-observer oriented model of the PEM fuel cell is detailed. Based on this model, theoretical and practical observability issues are discussed. This analysis shows that convection phenomena can be considered negligible from the dynamic point of view; this leads to a reduced model. Finally a state-observer enables the estimation of the inner partial pressure of the cathode by using only the current and voltage measurements. This proposed model-based approach has been successfully tested on a PEM fuel cell simulator using a set of possible fault scenarios.