BACKGROUND: Despite the explosive increase in interest regarding Robot-Assisted Walking Training (RAWT) for stroke patients, very few studies have divided groups according to the severity levels of patients and conducted studies on the effects of RAWT. OBJECTIVE: The purpose of this study was to present a definite basis for physical therapy using the robot-assisted walking device through a more detailed comparison and analysis and to select the optimal target of RAWT. METHODS: This study was designed as a prospective and randomized controlled trial to investigate the effect of RAWT on balance, motor function, and Activities of Daily Living (ADL) depending on severity levels in stroke patients. 100 participants were randomly divided into study and control groups in equal numbers. The study group was 49 and the control group was 47. One from the study group and three from the control group were eliminated. The study period is four weeks in total, and RAWT is performed five times a week for 40 minutes only for study group. During the same period, all group members had 30 minutes of Conventional Physiotherapy (CP) five times a week. RESULTS: The results of this study clearly confirmed that RAWT combined with CP produces more significant improvement in patients with stroke than the CP alone. And they indicated that RAWT had a more considerable effect in the poor or fair trunk control group for trunk balance and in the high fall risk group for balance. In motor function, RAWT showed its value in the severe and marked motor impairment group. The total or severe dependence group in ADL experienced more improvements for RAWT. CONCLUSION: This study can be concluded that the lower the level of physical functions, the more effective it responds to RAWT. As demonstrated in the results of this study, the potential of current robotic technology appears to be greatest at very low functional levels of stroke patients. Patients with low functional levels among stroke patients may benefit from robot rehabilitation.