The effects of training with isokinetic maximal voluntary knee extensions were studied in stroke patients. Two groups of 10 patients each trained twice a week for 6 weeks. One group trained exclusively eccentric movements and the other exclusively concentric movements. The effects were evaluated from the following tests before and after the training period. The maximal voluntary strength in concentric and eccentric actions of the knee extensor and flexor muscles was recorded together with surface electromyography at constant velocities of 60, 120, and 180 deg.s-1 on three different days. The body weight distribution on the legs while rising and sitting down was measured with two force plates. The self-selected and maximal walking speeds and the swing to stride ratio of the paretic leg were measured. After the training period, the knee extensor strength had increased in eccentric and concentric actions in both groups (p < .05). The eccentric and the concentric strength in the paretic leg relative to that of the nonparetic leg increased in the eccentrically trained group (p < .05) but not in the concentrically trained group. The restraint of the antagonistic muscles in concentric movements increased after concentric (p < .05) but not eccentric training. A nearly symmetrical body weight distribution on the legs in rising from a sitting position was noted after eccentric (p < .05) but not concentric training. Changes in walking variables were not significantly different between the groups. Eccentric knee extensor training was thus found to have some advantages as compared to concentric training in stroke patients.