In hemiplegic gait the paretic lower limb provides less muscle power and shows a briefer stance compared with the unaffected limb. Yet, a longer stance and a higher power can be obtained from the paretic lower limb if gait speed is increased. This supports the existence of a ‘learned non-use’ phenomenon, similar to that underlying some asymmetric impairments of the motion of the eyes and of the upper limbs. Crouch gait (CG) (bent-hip bent-knee, about 30° minimum knee flexion) might be an effective form of ‘forced-use’ treatment of the paretic lower limb. It is not known whether it also stimulates a more symmetric muscle power output. Gait analysis on a force treadmill was carried out in 12 healthy adults and seven hemiplegic patients (1–127 months after stroke, median: 1.6). Speed was imposed at 0.3 m/s. Step length and single and double stance times, sagittal joint rotations, peak positive power, and work in extension of the hip, knee, and ankle (plantar flexion), and surface electromyography (sEMG) area from extensor muscles during the generation of power were measured on either side during both erect and crouch walking. Significance was set at P less than 0.05; corrections for multiplicity were applied. Patients, compared with healthy controls, adopted in both gait modalities and on both sides a shorter step length (61–84%) as well as a shorter stance (76–90%) and swing (63–83%) time. As a rule, they also provided a higher muscular work (median: 137%, range: 77–250%) paralleled by a greater sEMG area (median: 174%, range: 75–185%). In erect gait, the generation of peak extensor power across hip, knee, and ankle joints was in general lower (83–90%) from the paretic limb and higher (98–165%) from the unaffected limb compared with control values. In CG, peak power generation across the three lower limb joints was invariably higher in hemiparetic patients: 107–177% from the paretic limb and 114–231% from the unaffected limb. When gait shifted from erect to crouch, only for hemiplegic patients, at the hip, the paretic/unaffected ratio increased significantly. For peak power, work, sEMG area, and joint rotation, the paretic/unaffected ratio increased from 55 to 85%, 56 to 72%, 68 to 91%, and 67 to 93%, respectively. CG appears to be an effective form of forced-use exercise eliciting more power and work from the paretic lower limb muscles sustained by a greater neural drive. It also seems effective in forcing a more symmetric power and work from the hip extensor muscles, but neither from the knee nor the ankle.