A novel alarm-system design strategy, which takes full advantage of the inherent hardware and spatial redundancy in a process network, is proposed in this paper. Specifically, systematic procedures have been developed to identify independent methods for evaluating any alarm variable in the process and to synthesize corresponding alarm generation logic. In order to implement this logic, the error models in data reconciliation and the formulas for evaluating conditional probabilities of type I and II mistakes have also been derived. The results of simulation studies show that it is indeed superior to any of the existing design techniques. This is because the resulting alarm system is appropriately tailored to minimize the expected loss. More importantly, it is robust in the sense that the system performs satisfactorily even under the influence of various sensor malfunctions.