In this study, we investigated the aspects of semantic processing and phonological retention during repetition exercises in Chinese advanced learners of Japanese. Working memory (WM) capacity was manipulated as an individual factor. Participants repeated sentences under three experimental conditions: 0-, 3-, and 5-second delays. The results show that: (a) The semantic processing performance of participants with larger WM capacity did not differ depending on the starting time of repeating, while the performance of participants with smaller WM capacity was better when the 3- and 5-second delays were imposed than during the 0-second delay. (b) The phonological retention performance of participants, regardless of WM capacity, was best when the 5-second delay was imposed, but poorest following a 0-second delay. These results suggest that when learners perform repetition exercises, the efficiency of semantic processing differs depending on the WM capacity. However, regardless of WM capacity, learners perform semantic processing first, followed by phonological retention.