One way to provide powered lower limb prostheses with greater adaptability to a wearer's intent is to use a neural signal to provide feedforward control of prosthesis mechanics. We designed and tested the feasibility of an experimental powered ankle-foot prosthesis that uses pneumatic artificial muscles and proportional myoelectric control to vary ankle mechanics during walking. The force output of the artificial plantar flexor muscles was directly proportional to the subject's residual gastrocnemius muscle activity. The maximum force generated by a pair of artificial muscles fixed at nominal length was 3513 N. The maximum planter flexion torque that could be generated during walking was 176 Nm. The force bandwidth of the pneumatic artificial muscles was 2 Hz. The electromechanical delay was 33 ms, the time to peak tension was 48 ms, and the half relaxation time was 50 ms. We used two artificial muscles as dorsiflexors and two artificial muscles as plantar flexors. The prosthetic ankle had 25 deg of dorsiflexion and 35 deg of plantar flexion with the artificial muscles uninflated. The intent of the device was not to create a commercially viable prosthesis but to have a laboratory prototype to test principles of locomotor adaptation and biomechanics. We recruited one unilateral transtibial amputee to walk on a treadmill at 1.0 m/s while wearing the powered prosthesis. We recorded muscle activity within the subject's prescribed prosthetic socket using surface electrodes. The controller was active throughout the entire gait cycle and did not rely on detection of gait phases. The amputee subject quickly adapted to the powered prosthesis and walked with a functional gait. The subject generated peak ankle power at push off that was similar between amputated and prosthetic sides. Our results suggest that amputees can use their residual muscles for proportional myoelectric control to alter prosthetic mechanics during walking.