It has been observed that in languages with dominant-recessive harmony systems, while either roots or suffixes can be dominant (i.e., they can trigger a phonological change in their surrounding elements), prefixes are always recessive (Clements 2000, Bakovi´c 2000, Casali 2003 a.o.). While the ‘no dominant prefixes’ generalization is acknowledged in most studies of dominant harmony systems, there is, to our knowledge, no systematic explanation for the pattern. We propose that the generalization is epiphenomenal, a special case (when it holds) of a broader generalization that phonological derivations of morphologically complex words proceed in stages, which may correspond to syntactic phases. Specifically, following Newell 2008, Fenger 2020 a.o., we propose that elements in the first domain may have unrestricted phonological interactions, including structure-changing operations, but that beyond the first phase/domain, phonological operations may not change material that is fixed on the first cycle. For vowel harmony, this predicts that ‘low’ affixes may be dominant (root-altering) or recessive, but that ‘high’ affixes that participate in vowel harmony (contra Fábregas & Krämer 2020) may only be recessive. Since it is independently established that prefixes tend to represent syntactically higher morphemes than suffixes (Julien 2002), ‘no dominant prefixes’ as a trend falls out as a special case.