Phonological word-final vowel shortening has been reported for some words in Tokyo Japanese. This phenomenon typically occurs with a long vowel following a heavy syllable. In search of a phonetic motivation for the phenomenon, we investigated durational variations in short and long vowels in phrase-medial and utterance-final positions. A production study revealed that phrase-medially V:2 was shorter than V:1 in CV:1CV:2 words, as well as shorter than V:2 in CV1CV:2 words. Nevertheless, the durational distributions of V:2 in phrase-medial CV:CV:2 words and that of phrase-medial short vowels were distinct. The durational distribution of V:2 in phrase-medial CV:CV:2 words, however, overlapped substantially with that of utterance-final short vowels, due to utterance-final lengthening. A perception study indicated that the phonological length of V:2 in phrase-medial CV:CV:2 words was misidentified more often than vowels in other positions, short or long. The results were attributed to two interacting factors: (1) a substantial overlap in durational distributions of V:2 in CV:1CV:2 words and short vowels across phrase-medial and utterance-final positions (this overlap may lead to confusion if the listener finds it difficult to process the phrase-position dependency of vowel duration); and (2) the low frequency of occurrence of the long vowel.