The vowels /i/, /a/, and /u/ were synthesized using a Nova minicomputer described earlier [G. Herman, J. Acoust. Soc. Am.61, S68(A) (1977)]. Six normally hearing subjects made similarity judgments of the vowel stimuli which were constructed with the three formant frequencies published earlier for each vowel. The vowels had a rise‐decay time of 20 msec. The formant structures of the vowels varied for four experimental conditions: (1) all three formants present; (2)F1andF2present,F3absent; (3)F1andF3present,F2absent; and (4)F2andF3present,F1absent. Vowel durations of 200 and 300 msec were used for each condition. Twenty‐four experimental conditions resulted and 576 stimulus pairs were used. Subjects listened at MCL and used an equal‐appearing interval scale to make their similarity judgments. Ratings were converted to similarity matrices and submitted to INDSCAL to find perceptual dimensions. Features retrieved related totongue height, tension, andadvancement, withdurationhaving less importance.