Many phonological processes display positional differences. In prominent positions, features are likely to be contrastive and marked structures are often allowed to occur. In contrast, in non-prominent positions, features are likely to be neutralized and marked structures are generally prohibited. A well-known example is the coda/onset asymmetry in obstruent voicing : in many languages, voice feature of an obstruent is contrastive only in the onset. For the Optimality-Theoretic analysis of such positional asymmetric facts, there have been two lines of approaches. The one adopts Positional Faithfulness constraints as the main device (Steriade 1995 : Jun 1995 : Beckman 1997, 1998 : and Kirchner 1998), whereas the other adopts Positional Markedness constraints (Flemming 1995 : Steriade 1997 : Zoll 1996, 1998). PM and PF analysis, a context-free markedness constraint is ranked between two faithfulness constraints. Given that PM and PF constraints aim to explain the same type of facts and the analyses adopting them proceed in a similar way, it will be redundant if we retain the both types of position-specific constraints in a grammar. The purpose of the present study is to determine what is the proper formal mechanism for positional asymmetric facts in general. We compare PM and PF approaches while discussing how they treat each of positional asymmetric facts observed in the literature. Our conclusion is that PM is in general superior to PF although PF is necessarily needed at least in accounting for directionality of assimilation.