To avoid harm caused by falsified medicines, we aimed to devise a method to identify falsified medicines in Japan using visual observation among medicines obtained from personal import agency websites, which are the main conduits through which falsified medicines are obtained. We recorded details regarding the information provided on personal import agency websites used to purchase medicines, the outer package received, the customs declaration description, and the product appearance for 212 samples of medicines obtained through personal import via the Internet. We investigated the relationship between each observed item and the rate of falsified medicines. We developed a classification and prediction model to identify falsified medicines using items that could be visually observed. The results showed that the rate of falsified medicines was significantly higher for websites that did not contain identifying information such as the name and address of the contact or import agency, as well as for products that did not contain the name and address of the manufacturer, indicating that these items may be useful in the identification of falsified medicines. In the prediction model constructed, we extracted features such as the country of dispatch and address of the import agency, and a prediction model was created to identify falsified medicines and websites selling these medicines. Careful observation of the identified features and use of our prediction model will help to prevent harm owing to the use of falsified medicines.