This paper studies the different ways in which ancient Sanskrit grammarians presented nominal declensions. Based on twelve Sanskrit grammars (from Pāṇini’s Aṣṭādhyāyī up to Bhaṭṭoji Dīkṣita’s Siddhāntakaumudī) classified into three categories (“wholly generative” grammars, “partly generative” grammars and “pedagogical” grammars), it shows that though nominal paradigmatic sets were known from and used by various scholars – among whom grammarians – at a relatively early date, it is only with pedagogical grammars that they really enter the scene of grammatical description, i.e. as “official” grammatical or language teaching tools.