The present study reports native speakers' impressions of JFL learners' utterances before and after shadowing/repeating training. Evaluation was also done for prosodically modified synthesized stimuli in order to examine the crucial prosodic cues. The results suggest that both durational patterns and pitch patterns are important for the utterances to be heard as natural Japanese, but durational patterns may be more important. Moreover, shadowing training appears to improve mora-timed rhythm. The results of the present study could provide useful suggestions for developing pronunciation training for Japanese pronunciation and speech education.