The scientific implications and practical applications of the Stein estimator approach for estimating true scores from observed scores are of potentially great importance. The conceptual complexity is not much greater than that required for more conventional regression models. The empirical Bayesian aspect allows the examiner to incorporate his/her own degree of prior information about selected examinees. This approach allows for a more accurate estimation of true scores, with the corollary of using fewer test items to achieve those true score estimates. Efron and Morris make the point that there is little penalty for using the rules discussed here because they cannot give larger total mean squared error than the maximum likelihood estimate. This assurance may be a sufficient reason for more careful examination of the utility of the Stein estimator and its limited translation estimators as they apply to behavioral and social science research. A numerical example is given of prediction of a season's batting averages on the basis of a sample of about 10 percent of the times at bat. (Author/CTM)