Processing instruction (PI) and skill-specificity may not be compatible with each other in SLA. Though VanPatten (2002a) has argued that PI was never intended to refer to comprehension versus production, PI and skill-specificity make opposite claims at least in terms of the usefulness of output practice. This paper discusses major arguments and empirical studies of each side and suggests following three directions to take for future research. First, further research on the effect of PI on various cross-linguistic structures and better operationalization of traditional instruction would help solve the problem. Second, both sides should differentiate two kinds of knowledge, declarative and procedural knowledge in Anderson`s framework (1993) and do more empirical studies on procedural knowledge. Finally, both sides should take into consideration the auto-input function of output suggested in the literature (Ellis, 1994; Levelt, 1989, Levelt, Roelofs, & Meyer, 1999; Platt and MacWhinney, 1983).