In the last five ctecades, research on Machine Translation of human languages, has produced many translation systems operating on a wide range of languages, text types and subject domains. Despite moderate success, there have been, only a few systems to answer the industrial demand. In this paper we discuss criteria that determine the feasibility of Machine Translation, and propose optimum points at which Machine Translation can be better implemented. Structure of a prototype system for Turkish language is also included in discussions, which is based on the results of Turkish-English Newsreport Translation project.