In recent years, adaptation to walking on a split-belt treadmill has been used as a common paradigm to explore how humans and other vertebrates adapt to walking in an asymmetric environment. When people walk on a split-belt treadmill, they initially walk with an asymmetry characterized by steps of unequal length, then gradually adopt a symmetric walking pattern characterized by steps of the same length. Although it has been proposed that the adoption of a symmetric walking pattern may result from a desired to maintain dynamic balance, this has yet to be investigated. Therefore, we used a model and experiment to explore how dynamic balance is influenced by spatiotemporal asymmetry during walking on a split-belt treadmill. The model was constructed based on a two-state exponential model which updates foot placement at each step based on the asymmetry from the previous. In the experiment, participants adapted to walking on a split-belt treadmill with the left belt moving at 1.5 m/s and the right belt moving at 0.5 m/s. Retroreflective markers were used to measure lower extremity kinematics and to compute spatiotemporal metrics such as step length, dynamic margins of stability, and time-to-contact. We found that people showed greater margins of stability on the fast belt than on the slow belt during adaptation. This was accomplished by adjusting the angle of the leading limb, which determined the placement of their feet relative to the body's center of mass. The results also showed greater times to contact on the slow belt than on the fast belt during adaptation and the difference between the fast and slow belt decayed during adaptation. These experimental results were consistent with the simulation using the computational model. Our results help to improve our understanding of the role of biomechanics in driving adaptive changes to coordination when walking in novel environments.