This paper presents a formal model of metacognitive reasoning in intelligent systems (IS). The proposed model was named fM2 and uses predicate logic to represent a cycle of reasoning about failures generated in reasoning tasks in an IS. fM2 has mechanisms such as introspective monitoring and meta-level control to perform metacognitive reasoning. fM2 was implemented and validated on an intelligent tutoring system named FUNPRO. The performance metrics of FUNPRO indicate the capacity of fM2 to drastically decrease the reasoning failures produced in the recommendations of FUNPRO. Thus, this paper demonstrates the efficacy of fM2 as a valid tool to improve the performance of the reasoning processes of IS.