This paper reports results of a series of vowel training studies. Study 1 trained two groups of Japanese learners of English on American English vowels and examined the effects of training set sizes (nine vs three more difficult vowels); Study 2 trained Korean learners of English and examined the efficiency of training protocols using both nine- and three-vowel sets. Study 3 compared the Japanese and Korean results on the untrained materials. Results suggested following: 1) vowel training works best when a large set of vowels, rather than a subset, is used; 2) training focusing on a smaller yet difficult vowels may have detrimental effects on later learning; and 3) improvement due to training on nonsense words may or may not carry over to untrained real word materials, possibly due to an interaction between native and non-native phonology.