This paper deals with the data of dormitory students in National Institute of Technology, Kagoshima College to demonstrate the existence of peer effects in academic performance. The data have unique advantages to avoid the difficulties of the self-selection problem and reflection problem. The data shows freshmen’s academic performance and previous year’s junior high school records, and roommate’s previous year’s academic performance for using an instrumental variable method. The results of my findings suggest that peer’s academic performance does not have any effects on freshmen’s. In spite of considering the asymmetric relationship between roommates, self-selection bias when choosing subjects, and nonlinearities of effects, there is no significance in any models.