The purpose of this work is to achieve better quality Machine Translation compared to the usual sentence-based Machine Translation. Discourse', rather than sentence' is used here as the unit of analysis. One principal technique in this approach is the resolution of anaphora in the source text to a Machine Translation system, which is the theme of this paper. This avenue, i.e. anaphora resolution, which is part of our work, is the focus of this discussion in the context of Machine Translation. The paper, therefore, discusses different kinds of anaphora (i.e. noun-phrase and verb-phrase anaphora) and suggests methods which help in their resolution. These methods include extraction of knowledge in one step and its use in the succeeding step(s), thereby indicating the usefulness of knowledge as a tool in this research. A brief discussion of this approach, which is being implemented as part of a Machine Translation system is presented in this paper.