Alarm systems provide an interface between operators and the machinery in a chemical plant because they allow abnormalities or faults caused by operators to be detected at an early stage. Alarms should be used to enable operators to diagnose faults and plan countermeasures, and nuisance alarms should be eliminated. We propose the use of a selection algorithm of sets of pairs of alarm variables and their signs. The signs mean the upper or lower limits of the alarm variables. The selected sets of pairs are theoretically guaranteed to be able to qualitatively distinguish all assumed faults. We propose using a two-layer cause-effect model for the algorithm, which represents the cause and effect relationship between state variables based on the topology of a plant. This model is applied to a simple process. The simulation results illustrate the usefulness of our method.