The spectral center of gravity refers to a listener’s averaging of frequency and intensity components when formant peaks in a speechlike signal are separated by 3.5 Bark units or less. In this paper a total of 18 synthetic vowels whose spectra approximated /ae/ or /inverted vee/ were generated digitally; each stimulus contained the first 40 harmonics of a 100‐Hz fundamental. Nine spectra contained three formants, while the balance contained only two. Subjects with normal hearing and mild high‐frequency hearing loss above 3000 Hz were instructed to identify synthetic vowels as either /ae/ or /inverted vee/ as F2 frequency was varied between nine different values in 100‐Hz steps for both the two‐formant and three‐formant stimuli. Probit analysis indicated that the normal‐hearing subjects identified stimuli more consistently than the mildly hearing‐impaired listeners across F2 frequencies for three‐formant than for two‐formant spectra. The F2 frequency corresponding to the perceived increase in vowel frontness occurred at a lower frequency for normal‐hearing listeners. Auditory filter bandwidth was negatively correlated with the F3−F2 Bark difference. Results suggest that spectral averaging may help listeners disambiguate confusing speech signals.