The Mixed Model Repeated Measures (MMRM) analysis has gradually been applied to analyze the clinical trial data in some therapeutic area. However, it is well known that the analysis results would depend on the specifications of the model. For example, what effects should be included in the model and should they be set to be fixed or random? How about the error correlation structure? Should the unstructured correlation always be considered first? What to do when it can not converge? Regarding the baseline data, should we use them as an independent or dependent variable? In this presentation, the issues and cautions of selecting suitable specifications for the model, with an emphasis on the correlation structure and the detailed data format, will be discussed.