Recently, there are many chances for university students to participate in online activities such as lecture or job interview. This study focused on online group discussion (GD), and proposed the random forest-based method to estimate listeners’ emotion during GD. In this study, we estimated listeners’ emotions with multimodal information such as movie and voice during GD, body acceleration and angular velocity. As a result, we did not obtain accurate estimation results, but an effectiveness of behavior information measured by acceleration and angular velocity sensors was implied.