James-Stein estimators are shown to be dominated by positive-part versions when estimating the mean of a p-variate normal distribution (p [greater-or-equal, slanted] 3), where the covariance matrix is known up to a constant. A Monte Carlo study is undertaken to compare their risks.