The plant fault detection and diagnosis system detects a fault in a plant under feedback-controlled operation and diagnoses the malfunctioning actuator/sensor as a fault cause by monitoring the manipulated variables and the process variables of the plant. This system supports decision-making by the plant operators. Although the difference in behavior between an actual process and its process model can be a basis for diagnosis, modeling error may lead to the wrong diagnosis. This difference consists of both a malfunction effect and a modeling error effect. An earlier study presented a diagnosis system that uses a filter to remove the difference due to modeling error effect from the observed difference and diagnoses based only on the difference due to malfunction effect. However, this filter removes not only modeling error effect but also the malfunction effect if it exists in the space for removal by that filter. So, this filter is a double-edged sword for a diagnosis system, since a serious malfunction may go undetected. Another system dealing with a limited modeling error that diagnoses actuator/sensor fault separately has also been presented. However, this system has the disadvantage that potential faults between actuator and sensor cannot be distinguished. This study presents a novel system in which modeling error and actuator/sensor fault are diagnosed by the same method, and which also allows reliability to be added as a potential cause of fault.