Humans conserve energy during walking using an inverted pendulum mechanism during single-limb support. The step-to-step transition requires substantial muscle-tendon mechanical work by the trailing lower limb on the center of mass. Currently we have limited understanding of how the ankle, knee, and hip contribute to center of mass mechanical work and overall metabolic energy expenditure during human walking. I used lightweight bilateral robotic exoskeletons powered with artificial pneumatic muscles to replace ankle joint mechanical work and examine changes in users’ metabolic energy consumption during walking. First I studied walking on level ground at preferred step length. Ankle exoskeletons replaced 22% of the lower-limb joint positive mechanical power and users saved 10% net metabolic power. For each 1 J of exoskeleton mechanical assistance subjects saved ~1.6 J of metabolic energy. The ‘apparent efficiency’ of ankle joint muscle-tendon positive mechanical work (0.61) is much higher than for isolated muscle positive mechanical work (0.25). This suggests that Achilles tendon contributes -60% of the ankle joint positive work, leaving -40% to active muscle fiber shortening. The ankle joint, therefore, performs 35% of the total lower-limb positive mechanical work but consumes only 17%-20% of the total metabolic energy during level walking at the preferred step length. In the next two experiments I used the powered exoskeletons to study ankle mechanics and energetics during walking with increasing step lengths and on increasing uphill inclines. Ankle joint ‘apparent efficiency’ decreased for walking with longer steps (0.39 at 140% of preferred step length) and on uphill gradients (0.38 at 15% grade). Thus, even when the demand for external positive mechanical work is high, the Achilles tendon still delivers 34% or more of the ankle joint positive work during walking. Overall these studies demonstrate that Achilles tendon elastic energy storage and return allows the ankle joint to perform positive mechanical work with very little metabolic cost. In contrast, knee and hip joint positive mechanical work performed by actively shortening muscle fibers likely exacts a much higher metabolic cost. Orthotic devices designed to reduce metabolic energy consumption during walking should target less efficient proximal joints (e.g. hip or knee).