In this paper it will be argued that for an adequate derivation of the domain of application of French liaison, direct reference to syntactic structure via domain-c-command (as proposed in Kaisse 1985) does not yield correct results. An analysis where an intermediate, prosodic structure (as in Selkirk 1986) is assumed does considerably better. Nevertheless, it will be shown that not the small phonological phrase (as argued in Selkirk 1986) but rather the prosodic word is the domain of obligatory liaison. In the small phonological phrase, liaison occurs variably. Liaison also occurs in a third type of contexts, namely after phrasal heads. This type of context can be mapped into a third type of prosodic constituent, the maximal phonological phrase. The final proposal will be that for an adequate account of French liaison a multilayered prosodic constituent structure must be assumed. This assumption is corroborated by real speech data from the Orleans corpus (see Blanc and Biggs 1972). © 1990, Walter de Gruyter.