Many process mining approaches have been published, which would be helpful to process modeling, but little attempt has been made for role mining used for organizational modeling. In this paper, a method of role mining using event logs as input is introduced. We adopt cosine metrics as the similarity measurement to deduce flat organizational model by clustering people base on their similar skills. In contrast to traditional techniques (e.g., questionnaire), our approach is objective and efficient. This paper also points out some challenges for future research.