Variability of step kinematics has been associated with falls by older adults. However, between-study differences with regard to the number of steps used to compute variability have varied by an order of magnitude. If the number of steps used to compute variability is too low there is the potential for a statistically spurious outcome. On the other hand, for subjects with mobility impairments a protocol requiring too many steps to estimate variability imposes an unnecessary burden on the subjects. We have determined the minimum number of steps needed to estimate the variability of spatial and temporal step kinematics. More than 700 steps were collected during level walking on an instrumented treadmill. Accurate estimation of step kinematic variability required at least 400 steps. The increased error in estimating the mean and standard deviations of the step kinematic variables with too few steps can impose an experimental cost with regard to statistical design considerations. The extent to which translation of these results can be made to the variability of spatial and temporal step kinematics collected during over-ground walking awaits further research.