This study makes use of a listening for mispronunciation task to examine how native English listeners perceive sentences produced by non-native speakers. The effects of target predictability and degree of foreign accent were investigated. Native and non-native speakers produced English sentences containing mispronunciation. Mispronunciations (MPs) were constructed by changing the initial phoneme of target words by a single distinctive feature along the dimensions of voicing, place, or manner. Results showed that listeners (a) were more accurate and faster in detecting MPs produced by native than non-native speakers, (b) were more accurate and faster in detecting MPs in predictable than unpredictable sentences, and (3) were more accurate in detecting MPs produced by non-native speakers with milder accents, as compared to heavier accents. These findings suggest that listening to fairly intelligible but accented speech requires increased processing effort--possibly because of subtle differences in intelligibility and increased variability characteristic of non-native speech.