Here we report the result of series of perceptual categorization experiments with the TIMIT Acoustic Phonetic Speech Corpus. The goal of these experiments was to identify primary acoustic and linguistic factors effecting the perception of naturally produced vowels. Over a set of three experiments, vowel tokens were presented to listeners in the following five different types of context: (1) no context at all; (2) the vowel in CVC context; (3) the vowel in an 'stripped'-syllable context; (4) the vowel in one-syllable-word context; and (5) the vowel in two-syllable-word context. A primary result is that identification performance improved with greater amounts of context (ie. 66.3% no-context, 79.4% CVC, and 81.5% word for the one-syllable word tokens.) These results provide evidence in support of non-segmental theories of vowel representation. Also, an interaction in identification accuracy over the three experiments suggests differences in linguistic redundancy across the token sets.