One of the few characteristics shared by most of the proposed algorithms for vowel normalization is their utilization of the mean formant frequency values of the vowel system as a correction factor for individual vowels. This reflects the accepted view that the quality of a particular vowel depends less on the absolute formant values than on the relations between these values and the general range of the speaker's formants. While such approaches very satisfactorily reduce the dispersion within a single language, they are unsatisfactory for cross‐linguistic comparisons, since the means can differ drastically from language to language when the vowel systems are different. For example, the overall mean of theF2values of Danish, which has two sets of front vowels (rounded and unrounded) but only one set of back vowels, is significantly higher than that of English; moreover, the overall meanF1of Danish is over 100 Hz lower than that of English. Using overall means to normalize in these cases confounds phonetic differences with systemic differences. A method of comparing vowels across languages which avoids this problem will be discussed. [Work supported by NIH.]