This study examined the cognitive mechanism of shadowing in Japanese as a second language for advanced class of Japanese learners. The concrete aim of this study is to explore the relation between learners' memory span (including working memory capacity and phonological memory span) and sentence structure (i.e. SOV and OSV). Two shadowing conditions (simultaneous and delayed shadowing) and one repeating condition are adopted in the experiment. Three dependent variables are measured as subjects' performances: the fluency of oral reproduction, recognition accuracy of presented sentences, and reaction time (only in shadowing conditions). The main results were as follows: (a) the higher-span group demonstrated more fluency than the lower-span group, (b) higher fluency under the repeating condition was observed in OSV sentences, (c) recognition accuracy rates are higher both under the delayed shadowing condition for the higher-span group and under the repeating condition for the lower-span group. Moreover, the reaction time of SOV sentences was faster than that of OSV under the simultaneous shadowing condition. These results suggested a possibility that not only phonological processing but also semantic processing occurs during shadowing in Japanese as a second language. However, it seems that these two processing occurs in parallel only for the learners of higher-span of working memory.