Underground urban expressways are a possibility for solving many existing transportation-related problems, such as traffic congestion in high density areas and the division of neighborhoods due to elevated roadways. However, they may also pose high risks regarding traffic safety. Therefore, it is important for a pre-analysis of traffic safety to be made. This paper describes recent efforts to develop a driving simulation system, MOVIC-T4, for traffic safety analysis of underground urban expressways. In order to develop a small portable simulator, a small-sized motion-base with two-degrees-of-freedom is used to duplicate accel-eration cueing together with a head-mounted-display (HMD) for the visual system. An overview of this sys-tem is given and the reliability of driving data obtained from the experiments using MOVIC-T4 is discussed through a validation study using field driving data. The results of validation indicate that the perceived speed, distance headway, and physiological data in the simulator show the almost same trend as that in the real world, but larger decelerations tend to be produced in the simulator. Underground urban expressways are a possibility for solving many existing transportation-related problems, such as traffic congestion in high density areas and the division of neighborhoods due to traffic safety. Therefore, it is important for a pre-analysis of traffic safety to be made. This paper describes recent efforts to develop a driving simulation system, MOVIC-T4, for traffic safety analysis of underground urban expressways. In order to develop a small portable simulator, a small-sized motion-base with two-degrees-of-freedom is used to duplicate accel-eration cueing together with a head-mounted-display (HMD) for the visual system. An overview of this sys-tem is given and the reliability of driving data obtained from the experiments using MOVIC-T4 is discussed through a validation study using field driving data. The results of validation indicate that the perceived speed, distance headway, and physiological data in the simulator show the almost same trend as that in the real world, but larger decelerations tend to be produced in the simulator.