It is well-established that in Tokyo Japanese, high vowels /i, u/ become devoiced between voiceless obstruents (e.g., /kuʧi/ → [ku̥ʧi] ‘mouth’). However, there is an ongoing debate on the exact phonetic/phonological status of these devoiced vowels. Some studies argue that devoiced vowels are completely deleted (Ogasawara 2013; Vance 2008) while others argue that they only lose their voicing while fully retaining their supralaryngeal gestures (Funatsu & Fujimoto 2011; Varden 2010). The disagreement seems to stem in large part from methodological differences. Studies that argue for deletion rely primarily on acoustic data of contexts in which only one of the two high vowels are phonotactically legal (e.g., [ʦuki] ‘moon’; *[ʦiki]), while studies that argue against deletion rely primarily on acoustic and articulatory data of [k]-initial contexts in which both high vowels are phonotactically legal in Japanese (e.g., [kiʧi] ‘good fortune’; [kuʧi] ‘mouth’). This talk presents a series of acoustic and articulatory studies on Japanese high vowels that control for different phonotactic predictability levels to tease apart methodological vs. phonetic/phonological effects. Results consistently show that complete deletion is possible, but also reveal an articulatory-acoustic disconnect, where devoiced high vowels are articulatory present but acoustically absent in certain contexts. The disconnect, together with perceptual results, could provide insight into different mechanisms behind production and perception grammars.