A review of cross-validation shrinkage formulas is presented which focuses on the theoretical and practical problems in the use of various formulas. Practical guidelines for use of both formulas and empirical cross-validation are provided. A comparison of results using these formulas in a range of situations is then presented. The result of these comparisons indicate that one should use Cattin's formula to estimate cross-validated R, employing either Wherry or Olkin-Pratt estimates of the population R. If examination of predictor-criterion correlations has occurred prior to regression analysis, use empirical cross-validation, or adjust p to indicate the original number of variables examined. Double cross-validation is considered inefficient and unsatisfactory, and a cautionary remark concerning the functional number of predictors is presented. (Author/PN)