CAN (Controller Area Network) is a broadcast bus with priority based access to the medium which has become a de-facto standard for data transmissio- n in automotive applications. To prevent a defective node from perturbing the functioning of the whole system, for instance by repetitively sending error frames, the CAN protocol includes fault confinement mechanisms whose objectives are to detect permanent hardware dysfunctioning and to switch off defective nodes. In this study, we derive a Markovian analysis of these mechanisms and identify several shortages. New mechanisms that address theses problems are then proposed and we provide the algorithms for their implementation.