Fault attack represents one of the serious threat against Java Card security. It consists in physical perturbation of chip components that cause an unusual behavior in the execution of Java Card applications. This perturbation allows to introduce faults on Java Card appli- cation, with the aim to reveal a secret information that are stored in the card or grant an undesired authorization. This paper presents a method- ology to recognize the sensitive code to the fault attack in the Java Card applications. It is based on concept from textcategorization and machine learning.