Considers the problem of estimating the covariance matrix of a normal distribution. In very large samples the maximum likelihood estimator (MLE) is of course "best" in many respects. In small or moderate samples, however, it is not surprising to fi...