Industrial plant alarm systems form an essential part of the operator interfaces for automatically monitoring plant state deviations and for attracting the plant operator's attention to the changes that require his/her intervention. In this paper, we describe a new method for selecting alarm source signals using a cause-effect model. (A cause-effect model is an abnormality propagation diagram that shows the propagation process of abnormalities after a malfunction on the basis of the material and energy balances of the plant.) The derived sets of alarm source signals are theoretically guaranteed to be able to statistically identify all assumed malfunctions. To evaluate the sets of alarm source signals quantitatively, we developed a preference index using the reachability matrix of the cause-effect model. The index is confirmed by using the structured cause-effect model. The design method was applied to a simple reaction process for this case study, and the simulation results proved the usefulness of the proposed approach.