This study investigates the representations of words’ pronunciations in the mental lexicon, and how they are used in spoken word recognition by native (L1) and non-native (L2) listeners. We report two long-term repetition priming lexical decision experiments, each conducted with native English listeners, Dutch learners, and Spanish learners of English. We tested whether the listeners recognized word repetitions more quickly and/or more accurately when the first (‘prime’) and second (‘target’) occurrence of the word shared surface details (e.g., the voice of the speaker) compared to when they did not. If so, this suggests that listeners retain word tokens with their acoustic details in their memories (in the form of an exemplar). We found that in one of the experiments, L2 listeners relied more on exemplar representations than L1 listeners did. Specifically, larger exemplar effects arose for the non-native listeners in the experiment in which we included reduced pronunciation variants resulting from acoustic schwa reduction in English (e.g., balloon). Reduced pronunciation variants are highly difficult to process for non-native but not for native listeners. Our finding suggests that exemplars play a role in spoken word recognition under specific circumstances only, such as under challenging listening conditions for L2 learners.