Some of the most fundamental questions in linguistic theory concern grammatical architecture. Focusing on morphosyntax specifically, how many components are ‘morphosyntactic’ phenomena distributed over, and how do they divide up their labor? Answers differ: some versions of Distributed Morphology posit a rich postsyntactic morphological component, whereas Morphology-as-Syntax approaches sharply reduce its role. Against that backdrop, this article investigates theme vowels, which are often analyzed as purely morphological (nonsyntactic). A diagnostic is introduced for their derivational origin, based on the narrow-syntactic phenomenon of lexical selection (L-selection): if a theme vowel is L-selected (to the exclusion of others) by a higher head, it must be in the narrow syntax; otherwise, the test is inconclusive. The former situation obtains in Latin: in synthetic causatives, fac- ‘make’ can be immediately preceded by -ē/-e but no other theme vowel. An analysis is developed on which theme vowels are syntactic and hence L-selectable. Alternative analyses on which theme vowels realize dissociated nodes added postsyntactically fail empirically or become notational variants of the syntactic analysis, since they must give theme vowels narrow-syntactic featural ‘precursors’. Theme vowels, then, are syntactic, introduced by (External) Merge. Insofar as dissociated-node insertion can be replaced with Merge, suspicious theoretical duplications are avoided, in line with minimalist goals.