The purpose of the present study was to investigate the effectiveness of acute phase hybrid assistive limb (HAL) rehabilitation training for patients after stroke by measuring the difference in the severity of paralysis. Fifty-three acute stroke patients were enrolled in this prospective cohort study. HAL training was administered about twice per week, and the mean number of sessions was 3.9 ± 2.7. The walking training was performed on a treadmill with individually adjustable body weight support and speed and there was a 10-m walk test (10MWT) before and after each session. Assessment at baseline and at endpoint consisted of the Glasgow Coma Scale (GCS), Revised Hasegawa’s Dementia Scale (HDS-R), Brunnstrom stage (Brs), Functional Independence Measure (FIM), Barthel index (BI), and 10MWT. We measured these assessments at the first walking training session and at the end of the final training session without the HAL. To evaluate the feasibility of training with the HAL, the outcome measures of BI, FIM, and speed and number of steps of 10MWT were compared before and after training using a paired Wilcoxon’s signed-rank test in different Brs. Except for Brs IV, the Brs III or higher subgroups displayed significant amelioration in BI, and the Brs III subgroup displayed significant amelioration in FIM. The Brs V and VI subgroups displayed significant amelioration in 10-m walking speed and steps. In acute phase rehabilitation after stroke, it is thought that the HAL is more effective for patients with less lower-limb paralysis, such as Brs III or higher.