We consider confidence sets for the mean of a multivariate normal distribution with unknown covariance matrix of the form[sigma]2I. The coverage probability of the usual confidence set is shown to be improved asymptotically by centering at a shrinkage estimator.