This study analyzed Korean learners’ acquisition process of za-gyo-sounds and the actual process of acquisition using longitudinal spontaneous speech data of three Korean JSL learners. As the result, the following facts were found. First, error had a higher frequency of occurrence when the occurrence position of za-gyo-sounds was an initial rather than the middle. However, it is necessary to check the objectivity through further analysis. Second, as the error occurred regardless of the time of the investigation, za-gyo-sounds was difficult to acquire for Korean learners as the speed of acquisition was too slow. Third, the error of za-gyo-sounds appeared to be concentrated in particular vocabulary. This implies that pronunciation instruction of vocabularies with a high frequency of error can greatly reduce the error rate of za-gyo- sounds. Furthermore, it means that instructing and correcting a particular vocabulary item, depending on the learning topic and stage, can contribute to the improvement of the pronunciation capabilities of learners.