Objective: To assess whether a downhill walking training programme is more effective than the same amount of training applied uphill in chronic stroke survivors. Design: Randomized, single-blind study. Setting: Outpatient rehabilitation service. Methods: Thirty-eight adults with hemiplegia from stroke lasting more than three months were randomly allocated to one of the two groups: 'UP' - 45 minutes of physical therapy + 30 minutes of treadmill with 5% ascending slope; and 'DOWN' - 45 minutes of physical therapy + 30 minutes of treadmill with 5% descending slope. Both groups were treated 5 times a week for six weeks. Patients were evaluated before treatment, at the end of treatment and after three months. Outcome measures: Primary outcome measure was the number of patients showing an improvement in 6-minute walking test (6MWT) greater than 50 m. Secondary outcome measures were: (1) number of patients showing a clinically relevant improvement of gait speed during 10-m walking test (10mWT); (2) number of patients showing an improvement in timed up and go (TUG) greater than minimal detectable change. Results: Both groups had a significant improvement after treatment and at follow-up. At the end of treatment, compared to UP group, more patients in the DOWN group showed clinically significant improvements in primary and secondary outcomes (16/19 patients for 6MWT, 11/19 patients for 10mWT and 9/19 patients for TUG compared with 3/19, 4/19 and 2/19 patients, respectively, P < 0.01). At follow-up, results were similar except for 10mWT. Conclusions: In chronic stroke patients, downhill treadmill training produces a bigger effect than uphill training.