Alarm flooding is a significant problem in the process industries. To solve this problem, a scenario-based early warning system design methodology is proposed. It comprises three steps: (i) scenario identification: events are identified by HAZOP analysis, variables are allocated to the scenario-based group, and the variables states correlated to the scenarios are identified; (ii) model development: Bayesian network of all variables is learned from the process data, and the events nodes are appended according to expert knowledge to construct the Bayesian network model of a scenario-based early warning system; (iii) model implementation: the model is applied online to monitor process, the monitored variables continuously produce evidence, update the events probabilities, find the root causes, and give an events warning message together with the root cause to operators. The methodology implementation and salient points are explained with the help of an easy to follow simple case study.