The decompositional, non-lexicalist, approach to word-structure and the theory of roots have proven highly rewarding. One grey area, however, comes from Romance since, rather than roots, the word structure primitive appears to be the stem: root + ‘stem-formative/theme vowel’. Regardless, there have been perspicacious decompositional accounts of Italian, however these are still marred by the large number of morphological/item-specific irregularities, motivating arbitrary noun classes. Additionally, there are roots that do not inflect: consonant-final and vowelfinal forms when these are oxytonic or loanwords. Given these irregularities, previous analyses in Italian have included the use of lexical exceptions and class features. We challenge the use of class features in generating the attested patterns (and their exceptions). Instead, we propose a new categorisation of root-shapes, which, when combined with the exponents of nominal inflection, produce the correct surface pairings, as well as the non-alternating forms. In our analysis, there is no diacritic or special marking of lexical exceptions, all forms inflect regularly in accordance to their phonological shape. This requires the innovation of one new mechanism (Inhibition), but we back it up by showing that it leads to an unexpected beneficial prediction that solves a long-standing problem associated with Raddoppiamento Sintattico (RS).