Purpose: The purpose of the present study was to examine, in stroke patients, differences between backward walking training applied on a treadmill and the same training applied on the ground. Methods: Twenty seven stroke patients were divided into a treadmill backward walking group of 14 patients and a ground backward walking group of 13 subjects. Each group performed their respective training method for 8 weeks (15 min per day, 4 days a week). Walking ability was measured using a 10 m MWS (Maximal Walking Speed) test and the GAITRite system to examine changes in walking. Cadence, stridelength, step time, step length and symmetry index of the less affected side were measured to examine changes in stance phase of the lower extremity of the more affected side. Results: 10 m MWS, cadence, stride length, step time and step length of the less affected side significantly increased and symmetry index significantly decreased after training in both groups. The treadmill backward walking group experienced a significantly greater increase in step time and step length and a significantly greater decrease in symmetry index than the ground backward walking group. Conclusion: The two walking training methods were effective for improving stability in stance phase of the lower extremity of the more affected side, but the treadmill method was more effective. The present study is meaningful in that it analyzed the effects of backward walking training methods on walking and the differences of the training methods to provide information necessary for effective treatment of stroke patients.