The Joola languages form a quite homogenous cluster in southern Senegal and Northern Guinea Bissau. However, this homogeneity is more obvious for grammatical features than for the basic lexicon. Lexical counts (Carlton & Rand 1993) show that there is considerable variation in the stability of the supposedly most stable part of the lexicon, with figures ranging from 90% or more to less than 15%. The minute examination of this puzzling situation may be regarded as a laboratory experiment that might serve to account for the long recognized fact that NC languages in general seem to share more grammatical (i.e. typological) features than lexical cognates. In this talk, I presented lexical series pertaining to basic lexicon but showing many different forms. I also showed how homogeneous the principal grammatical features are (with most paradigms showing very similar contents: personal pronouns, noun class markers, verb extensions, etc.). Finally, I showed how the Joola languages tend to renew their lexical stock, more by internal means (semantic shifts, lexical derivation) than from external influence. The general message I wish to deliver is twofold: - first, a serious lexical comparison cannot be undertaken with short wordlists only. For example , the only reflex of the NC root *DI ‘to eat’ in Joola Keeraak is found in the word mʊ-rɩ-aay-am ‘food’, whereas the regular Keeraak word for ‘to eat’ is now -ɲɔɔfɔ. Thus, a superficail comparison between Joola lects would result in considering that the NC root *DI is not represented in Keeraak. - second, the Joola case could be a model for similar investigations in other parts of the Niger-Congo domain. In fact, phenomens like polysemy or semantic shifts have been globally overlooked in lexical comparison of African languages, probably because during decades there were so few data available. This is not case anymore.