The present paper discusses the problem of deriving quotational expressions (QEs)–a relatively unexplored field in the generative tradition–in the context of phase theory (PT). QEs are taken to be built within Narrow Syntax. While this fact is empirically well-motivated, it turns out to be conceptually problematic for cyclic derivations driven by feature makeup. Empirical data show that QEs represent a unique mixture of conflicting properties. On the one hand, they must be interpretable as both atomic and not. This extends to two readings of the embedding of enquotation, which are argued to follow from recursive and purely iterative, non-recursive interpretation. On the other hand, having no limits in length and scope, QEs are more challenging for PT than other non-compositional chunks. I argue that these effects unearth an important conceptual contrast. While they pose certain problems for the Minimalist approach, they naturally follow from the account of Narrow Syntax proposed by Jan-Wouter Zwart. In this regard the gap in applicability of each account to the discussed phenomenon sheds new light on the role of features in PT.