This paper proves the second order correctness of the stationary bootstrap procedure for normalized, multivariate sample mean of weakly dependent observations. Similar results are shown to hold also for more general vector valued statistics based on sample means.