One of the issues in Distributed Morphology is whether it is linear locality (Embick, 2010; Kastner, 2019) or the structural locality (Choi & Harley, 2019 a.o.) that matters in allomorph conditioning. Examining Japanese verbal inflection, I propose that (i) the paradigm involves theme vowels, contra previous analyses, and (ii) structural locality is necessary to capture the allomorphy of those theme vowels.