A model for the development of a corpus-driven spelling dictionary for the Bambara language is described. First, a list of about 4000 lexemes characterized by spelling variability is extracted from an electronic Bambara-French dictionary. At the next stage, a script is applied to determine the number of occurrences of each spelling variant in the Bambara Reference Corpus, separately for the entire Corpus (more than 11 million words) and for its disambiguated subcorpus (about 1.5 million words). Statistics on the diversity of sources and authors are also obtained automatically. The statistical data are then sorted manually into two lists of lexemes: those whose standard spelling can be established statistically, and those requiring evaluation by expert linguists. Some difficult cases are discussed in the paper. At the final stage, a representative expert commission will discuss all those lexemes for which statistical data alone do not suffice to define a standard spelling variant, before taking a final decision on each. The resulting Bambara spelling dictionary will be published electronically and on paper.