In current versions of prosodic phonology, phonological rules are believed to apply only to the constituents in the prosodic hierarchy, each of which is mapped from the syntax by a unique algorithm (or constraint ranking). Since the primitives of the theory are the levels of the hierarchy, the prediction is that rules applying to a domain of a given size will be influenced by a uniform set of factors, while rules applying to domains of different sizes will not necessarily share any particular properties. This paper shows that this way of partitioning the data obscures several observations of potential interest. At the Phonological Phrase level in particular, there is a considerable amount of cross-linguistic variation in terms of what kind of information rules refer to. A survey of phrasal rules shows that (a) apparently branching-sensitive rules belong to a larger subclass of rules whose domains vary from utterance to utterance, and that (b) these rules in turn have the characteristic properties of a different level of the hierarchy, the Intonational Phrase. In the alternative model developed in this paper, these correlations arise because there are (at least) two distinct stages of domain formation, which are ordered with respect to other PF operations and hence make reference to different kinds of information.