The present study investigated whether listeners perceptually map phonetic information tophonological feature categories or to phonemes. The test case is a phonological feature that occurs inmost of the world’s languages, namely vowel height, and its acoustic correlate, the first formant (F1).We first simulated vowel discrimination in virtual listeners who perceive speech sounds throughphonological features and virtual listeners who perceive through phonemes. The simulationsrevealed that feature listeners differed from phoneme listeners in their perceptual discriminationof F1 along a front-back boundary continuum as compared to a front (or back) continuum. The competing predictions of phoneme-based versus feature-based vowel discrimination wereexplicitly tested in real human listeners. The real listeners’ vowel discrimination did not resemble thesimulated phoneme listeners, and was compatible with that of the simulated feature listeners. Thefindings suggest that humans perceive vowel F1 through phonological feature categories like /high/and /mid/.