In the modern world, systems are becoming increasingly complex, consisting of large numbers of components and their failures. In order to monitor system performance and to detect faults and diagnose failures, sensors can be used. However, using sensors can increase the cost and weight of the system. Therefore, sensors need to be selected based on the information that they provide.In this paper, a sensor selection process is introduced based on a novel sensor performance metric. In this process, sensors are selected based on their ability to detect faults and diagnose failures of components in the system, as well as the severity of failure effects on system performance. A Bayesian Belief Network (BBN) is used to model the outputs of the sensors. Sensor reading evidence is introduced in the BBN to enable the component failures to be identified. A simple system example is used to illustrate the proposed approach.