Walking on a split-belt treadmill (each of the two belts running at a different speed) has been proposed as an experimental paradigm to investigate the flexibility of the neural control of gait and as a form of therapeutic exercise. However, the scarcity of dynamic investigations challenges the validity of the available findings. The aim of the present study was to investigate the dynamic asymmetries of lower limbs of healthy adults during adaptation to gait on a split-belt treadmill. Ten healthy adults walked on a split-belt treadmill mounted on force sensors, with belts running either at the same speed (‘tied’ condition) or at different speeds (‘split’ condition, 0.4 vs. 0.8 or 0.8 vs. 1.2 m/s). The sagittal power and work provided by ankle, knee and hip joints, joint rotations, muscle lengthening, and surface electromyography were recorded simultaneously. Various tied/split walking sequences were requested. In the split condition a marked asymmetry between the parameters recorded from each of the two lower limbs, in particular from the ankle joint, was recorded. The work provided by the ankle (the main engine of body propulsion) was 4.8 and 2.2 times higher (in the 0.4 vs. 0.8, and 0.8 vs. 1.2 m/s conditions, respectively) compared with the slower side, and 1.2 and 1.1 times higher compared with the same speed in the tied condition. Compared with overground gait in hemiplegia, split gait entails an opposite spatial and dynamic asymmetry. The faster leg mimics the paretic limb temporally, but the unimpaired limb from the spatial and dynamic point of view. These differences challenge the proposed protocols of split gait as forms of therapeutic exercise.