Jim Flege’s research on category formation has introduced or emphasized several key concepts, including equivalence classification and the distinction between new and similar phones. The research described in this paper addresses these concepts by investigating the role of acoustic variability in the formation of new categories as well as the extent to which this variability may hinder or help native and non-native listeners. A production study comparing Spanish-accented and native English vowels reveals a much greater degree of variability in nonnatives’ use of the English vowel space. Results from a subsequent training study where vowel variability was systematically manipulated, suggests that for the most easily maintained distinctions, learning benefited from the high-variability training paradigm. In contrast, for very difficult distinctions, advantages were found for training only with minimal variability (prototypes). Finally, results are presented from a lexical decision task in which English and Dutch listeners responded to native and Dutch-accented English. While Americans prefer native English speech, the Dutch prefer the Dutch-accented stimuli. In addition, Dutch listeners are less efficient in processing words containing sounds that do not occur in Dutch even when listening to a native English speaker.