In this paper we propose an error model following a generalized Poisson process and enabling us to consider both single and "bursty" transmission errors. We deal with the worst-case deadline failure probability (WCDFP) evaluation in CAN. Worst-case means that the message transmission takes the maximum possible time and each error introduces a retransmission with the maximum overhead. The motivation of this work is that, in practice, the number of errors occurring during a given time period can not always be bounded, especially in the applications we consider (CAN used as an in-vehicle network). A method for efficiently computing, for each message, the WCDFP is presented. This method is general in the sense that it enables us to consider all possible burst-size distributions. We applied our analysis to an industrial case-study to estimate its reliability.