Few years ago, RFID systems were only deployed in single site scenarios, usually with few readers. Nowadays, this kind of architecture is not suitable for the new business needs, such as data sharing with partners' companies. To meet these new requirements, a novel type of software, called RFID middleware has been designed for managing RFID devices and for processing the huge volume of generated raw data. Besides, current RFID systems are increasingly used in critical domains such as medical field or real-time processing domains. So, significant efforts are made to enhance the reliability of these systems. Despite that, more research is needed to meet the increasing dependa- bility requirements. In this article, we propose a probabilistic diagnosis algorithm (called "RFID diagAlgo") that identifies faulty components of an RFID system on the basis of a statistical analysis of their read results.