One assumption that has driven the development of phonological models is the idea that a single underlying representation (UR) exists in the mind such that talkers draw on it in both producing and perceiving speech. Data from a wide range of linguistic research suggest mismatches in phonetic and phonological abilities between perception and production which would be incompatible with the single UR model. The research reported here was specifically designed to test whether changes in perceptual categorization behavior would influence speech production abilities in a manner consistent with a single UR. Twenty native speakers of English were taught to categorically perceive a three way voicing contrast in synthetic bilabial stop consonants which varied only in VOT [after Pisoni, Alsin, Perey, and Hennessey, JEP: HPP 8(2), 297–314 (1982)]. Perception and production tests were administered following training. Subjects showed the ability, which improved with training, to categorically identify the three-way voicing contrast. Subsequent acoustic and perceptual analyses showed that they were unable to produce the contrast correctly, producing no difference, or manipulating acoustic variables other than VOT (vowel duration, vowel quality, nasalization). These results are explained in terms of separate expressive and receptive representations and the implications for perception/production mappings discussed.