In the evaluation of medical device applications, sponsors sometimes propose mixed non-inferiority margins based on both absolute difference and relative risk. Although the type I error rate may be controlled statistically, it however can result an unacceptable outcome from clinical perspective. Consequently, it comes to the point that the choice of non-inferiority margin at the design stage not only should be statistically sound, but also clinically relevant and acceptable. At this roundtable, we will discuss some issues concerning this topic.