Consensus is known to be a fundamental problem in fault-tolerant distributed systems. Solving this problem provides the means for distributed processes to agree on a single value. This, however, requires extra communication efforts. For some real-time communication networks such efforts may have undesirable performance implications due to their limited bandwidth. This is certainly the case with the Controller Area Network (CAN), which is widely used to support real-time systems. This paper shows how some underlying properties of CAN can be used to solve the consensus problem. The proposed consensus protocol tolerates the maximum number of process crashes, is efficient and flexible. The described solution is proved correct, its complexity is analysed and its performance is evaluated by simulation.