Safety-critical applications are now common where both digital and mechanical components are deployed, as in the automotive fields. The analysis of the dependability of such systems is a particularly complex task that mandates modeling capabilities in both the discrete and in the continuous domains. To tackle this problem a multi-level approach is presented here, which is based on abstract functional models to capture the behavior of the whole system, and on detailed structural models to cope with the details of system components. In this paper we describe how the interaction between the two levels of abstraction is managed to provide accurate analysis of the dependability of the whole system. In particular, the proposed technique is shown to be able to identify faults affecting the CAN network whose effects are most likely to be critical for vehicle's dynamic. Exploiting the information about the effects of these faults, they can then be further analyzed at the higher level of details.