In Japan, there have been reports of more than 600 recalls per year for drugs, medical devices, cosmetics and other quasi-drugs. The Japanese Ministry of Health, Labor and Welfare defined a classification system referring to that defined by the FDA, which has been used to classify all recall cases since 2000. Nevertheless, a problem in assigning the cases to these classifications has emerged. In this study, we analyzed the recall data on medicines and medical devices disclosed by Japanese authorities to determine the reason for such misclassifications and analyzed them based on a text mining technique, the dependency-link method.