One of the fundamental problems in designing non-inferiority trials is the specification of the non-inferiority margin. In the available regulatory guidelines or publications, only general principles regarding the specification of non-inferiority margin are discussed. In the published literature, all non-inferiority related papers either avoided discussing the margin specification, or else simply assumed the margins are given. The anticipated FDA guidance on non-inferiority is expected to discuss margin specification based on the two current approaches. These approaches of margin specifications involve strong assumptions and meta-analysis of available historic control study data, and are often controversial. In this talk, a different method is proposed for the specification of non-inferiority margin for a variety of endpoints including normal, binary, survival, and Poisson.