In this paper, characteristics of a runaway project are revealed based on combination of risk factors which appear in the project. Concretely, an association rule mining technique is applied with an actual questionnaire data to induce rules that associate combination of risk factors with runaway status of software projects. Furthermore, the induced rules are integrated and reduced based on a certain rule obtained from experts’ perception to simplify the representation of characteristics of a runaway project. Then, for confirming the effectiveness of this characterization, it is evaluated how many runaway projects in distinct data set were identified by the reduced rules. The result of the experiment suggested that the induced rules are effective to characterize runaway projects.