In the present paper we have established that the two groups of Iranian EFL learners have a perceptual representation of the vowel of American English that deviates substantially from that entertained by American native speakers. We have not tested the intelligibility of our EFL learners for American native listeners or for non-native users of English as a lingua franca. Afshar (2022: 97) has provisionally tested the intelligibility of the vowels by training automatic classifiers with the native AE vowel tokens, and then used the classifiers to identify the non-native tokens. It would be beyond the scope of the present paper to include this part of our research. As a rough indication the best classifier identified 90% of the native AE tokens correctly, using F1, F2 and duration as input parameters, against 57% correct for the two learner groups (with no difference between them). We have also shown that the perceptual representation and the production of the vowels are strongly isomorphic, so that it seems plausible to assume that the deviant articulation of the AE vowels by our learner groups stems from incorrect perceptual targets of these vowels, both in terms of their vowel quality (color, as determined by the formants) and in terms of their duration. It is impossible to decide, on the basis of our experiments, whether the incorrect perceptual representation of the AE vowels is actually the cause of their incorrect vowel production. Quite probably, a learner’s pronunciation will improve if the learner has a correct perceptual representation of the target sounds, so that they will hear when their vowel production is off-target, and will retry until there is a sufficient match between the token produced and the perceptual target.