The aim of this thesis is to describe how corpus-based Setswana dictionaries should be compiled. The challenge to the modern Setswana lexicographer is to compile very practical descriptive and user-friendly dictionaries. A detailed evaluation of existing Setswana dictionaries will be performed in terms of the macrostructural and microstructural aspects: <ul><li> Coverage of frequently used words. </li><li> Effective use of dictionary space. </li><li> Use of standard dictionary conventions. </li><li> Choice, ordering and composition of translation equivalent paradigms. </li></ul> The focus will be on material collection and corpus building. Informants will be used to compile an oral corpus of 100,000 tokens. All ethical requirements such as informed consent requirements (See Appendix 1) will be honoured. Since the text corpus is an organic corpus, thus not a designed corpus aimed at balance and representativeness, the oral corpus will be constructed in the same way i.e. only basic selection criteria: <ul><li> Mother tongue speakers of Setswana. </li><li> Adults (to be on a par with authors of the written sources in the text corpus). Age: ranging from 20-60 years. </li><li> Male and female. </li></ul> Critical analysis of all currently available Setswana dictionaries will be done with special reference to the dictionaries of Brown (1987) (SESD), Snyman, et al. (1990), Matumo (1993).(MSED), Kgasa (1976) (THAND) and Kgasa and Tsonope (1995).(THAN) In all these cases the strategy would be in terms of the theoretical criteria and best practices in terms of a broad theoretical survey of core aspects of dictionary compilation. Finally, the study will be concluded with an analysis of corpus integrity and stability of Setswana corpora based on the model introduced by Prinsloo and De Schryver (2001a).