Abstract: The existing self-reflection models based on natural language or diagram have not been able to meet the requirements of verification and analysis. The PEPA model is converted into Ordinary Differential Equations (ODEs). A dynamic self-reflection model based on Performance Evaluation Process Algebra (PEPA) which provides an insight into quantitative analysis besides reasoning is proposed to describe the process of selfreflection. Using a fluid flow approximation method to avoid a state-space explosion of Markov chains, a traditional solution method. The analysis results demonstrate that it is important to improve the efficiency of selfreflection by decreasing the latency time of monitor as well as shortening the length of instruction sequences.