Previous evidence from category goodness rating tasks has demonstrated that the phonetic reflexes of phonological categories have internal structure, such that some signals are better cues to phonological category membership than others. Puzzlingly, the best-rated exemplars of peripheral vowel categories have often been observed to be more peripheral than even hyperarticulated signals from natural speech, while non-peripheral vowel categories have often failed to demonstrate clear patterns of internal structure. The present study proposes that these puzzles can be explained by a conception of the phonology-phonetics interface whereby phonetic implementation is primarily driven by the maximization of contrast, rather than by particular targets in acoustic or articulatory space. Two experiments were conducted to test this hypothesis. In Experiment 1, native American English speakers rated the category goodness of exemplars of the peripheral [æ] category, and native Turkish speakers rated exemplars of the non-peripheral [y] category. Results confirmed the generalization that peripheral vowels demonstrate clearer patterns of internal structure than non-peripheral vowels. In Experiment 2, native American English speakers rated exemplars of the [æ] category that were extremely peripheral in the F1-F2 space—far more peripheral than even the most hyperarticulated productions from natural speech. Results generally supported an indefinite linear relationship between formants and goodness, consistent with a view of phonetic implementation based primarily on contrast rather than particular targets. I argue that this view is broadly compatible with existing approaches to the phonology-phonetics interface, but would involve certain modifications. I suggest paths for future research to investigate the implications of these modifications.