Mining is one of the most dangerous industries. Mine Safety and Health Administration (MSHA) maintains a database that records thousands of mining related accidents, injuries or illnesses every year with incident descriptions in narrative texts. How to uncover knowledge from these narrative texts is lacking. The goal of this study is to propose a new data mining methodology that incorporates or extends existing methods and is able to uncover useful information from massive amount of narrative texts in a streamline fashion. In our experimentation with data of 2008, we focus on 3 different types of common injuries and apply the new methodology to their narrative texts. Some interesting results are found that are worthy further investigations with the help of mining safety experts.