The need to initiate a step in order to recover balance could, in theory, be predicted by a static model based solely on displacement of the center of mass (COM) with respect to the base of support (BOS), or by a dynamic model based on the interaction between COM displacement and velocity. The purpose of this study was to determine whether the dynamic model provides better prediction than the static model regarding the need to step in response to moving-platform perturbation. The COM phase plane trajectories were determined for 10 healthy young adults for trials where the supporting platform was translated at three different acceleration levels in anterior and posterior directions. These trajectories were compared with the thresholds for step initiation predicted by the static and dynamic COM models. A single-link-plus-foot biomechanical model was employed to mathematically simulate termination of the COM movement, without stepping, using the measured platform acceleration as the input. An optimization routine was used to determine the stability boundaries in COM state space so as to establish the dynamic thresholds where a compensatory step must be initiated in order to recover balance. In the static model, the threshold for step initiation was reached if the COM was displaced beyond the BOS limits. The dynamic model showed substantially better accuracy than the static model in predicting the need to step in order to recover balance: 71% of all stepping responses predicted correctly by the dynamic model versus only 11% by the static model. These results support the proposition that the central nervous system must react to and control dynamic effects, i.e. COM velocity, as well as COM displacement in order to maintain stability with respect to the existing BOS without stepping.