Bobaljik (2012) observes on the basis of an impressive sample of languages that root suppletion is hardly ever conditioned by degree markers that do not form a word with the root. He calls this the Root Suppletion Generalization (RSG). If true, the generalization provides a possible argument for the lexicalist position: RSG can be seen as a consequence of the lexicalist architecture, where words are built pre-syntax, and therefore syntax cannot influence their shape (Williams 2007). Against this background, this paper discusses evidence (some of it presented already in Bobaljik’s work) that the RSG (when stated over words) is empirically (sometimes) too weak and (sometimes also) too strong. In view of these observations, I suggest a way in which all of these examples can be captured in ways that do not lend any support to lexicalism, simply because the word is not the relevant notion for blocking suppletion.