In Prosodic Phonology, domains for the application of phonological patterns are commonly modeled as a Prosodic Hierarchy. The theory predicts, among other things, that (i) prosodic domains cluster on a single universal set of domains (‘Clustering’), and (ii) no level of prosodic structure is skipped in the building of prosodic structure unless this is required by independently motivated higher ranking principles or constraints (‘Strict Succession’). In this paper, we demonstrate that if, as is standardly done, evidence is limited to lexically general phonological processes, some languages systematically violate the Strict Succession Prediction, evidencing no prosodic word domain, and some languages systematically violate the Clustering Prediction, evidencing more than one domain between the phonological phrase and the foot. We substantiate these claims by in-depth studies of phonological rule domains in Vietnamese (Austroasiatic) and Limbu (Sino-Tibetan). As an alternative to the Prosodic Hierarchy framework, we advocate a heuristic for cross-linguistic comparison in which prosodic domains are conceived of as language-particular, intrinsic and highly specific properties of individual phonological rules or constraints. This allows us to explore empirically the actual degree of variation to be encountered across prosodic systems. It turns out that the ‘word’ has no privileged or universal status in phonology, but only emerges through frequent reference of sound patterns to a given construction type in a given language.