This paper focused on the effects of prosody shadowing of Japanese sentences on the production of special Japanese phonemes by Chinese learners of Japanese. Twenty-nine students participated in a time-series experiment in which phonological short-term memory and the numbers of prosody shadowing trials were the independent variables, and the production accuracy of Japanese words and that of sentences were adopted as the dependent variables. The results showed that all students improved on the production of Japanese special phonemes regardless of the capacity of phonological short-term memory. In addition, the learners showed improvement after the fourth or fifth trials but not earlier, suggesting that this number of task repetition is necessary for short-term prosody shadowing training. These results were discussed in terms of appropriate phonological encoding and the construction of phonological representation in Japanese.