Recent studies have shown that accurate vowel category perception can be maintained at fundamental frequencies (fo) up to at least 880 Hz. In such cases, the typical first formant frequency (F1) of most vowels is exceeded by fo and the vocal tract transfer function is to a high degree undersampled. It seems therefore unlikely that common formant patterns are the main acoustic features used by listeners to recognize vowels at high fos. Here, we present results from multidimensional scaling (MDS) analyses calculated by averaging cochlea-scaled spectra across multiple steady-state vowels (N = 324; all 250 ms) produced by a female native German speaker at nine fos within a range of 220-880 Hz. All vowels (/i, y, e, ø, ɛ, a, o, u/) were recognized accurately by listeners (N = 20) in a previous study [Friedrichs et al., The phonological function of vowels is maintained at fundamental frequencies up to 880 Hz, J. Acoust. Soc. Am. 138, EL36-EL42 (2015)]. MDS reveals that with increasing fo, the vowel height dimension partially collapses, but the front/back distinction expands, thus allowing the vowels to be distinguished. This indicates that the perceptual space is reorganized and vowel height and frontness are being combined in a correlated way at higher fos.