Abstract: In order to monitor, control and optimize the overall chemical process, it is necessary to observe different kinds of variables/parameters. The problem of placing sensors with respect to different criteria, for instance, total cost, system reliability, system accuracy, is known as the problem of sensor network design. This paper studies the problem of non-redundant sensor network design for linear processes. Much research has been done for this problem, but no one of them considered the case that multiple sensors could be used to measure the same variable. Additionally, prior studies neither considered cost, system reliability and system accuracy simultaneously, nor were capable of guaranteeing a global optimal result. In this paper, based on the graph theory concepts, a MILP model is presented for designing the cost-optimal sensor network considering system reliability and system accuracy. The proposed model also consideres the case that multiple sensors could be used to measure the same variable. In order to get a global optimal result, the MILP model is solved by using branch- bound method with the existing software LINGO. The efficiency of the MILP model is demonstrated through an application on the steam-metering network.