Conversion equations that are based on linear regression are used widely to transform estimated breeding values (EBV) of sires for production, type, health, and management traits from the genetic base, scale, and units of measurement of an exporting country to that of an importing country. One of the major deficiencies of these regression equations is that the accuracy of converted EBV of elite sires and cows, which are of primary interest in genetic selection programs, is lower than that for average animals. In this study, it is shown both mathematically and in practical examples that the standard error (SE) of prediction of elite dairy sires can be much larger than for average sires. When more than 100 sires are used to develop conversion equations, the SE of prediction for elite AI sires is up to 10% larger than for an average sire, and, when 50 sires are used to develop conversion equations, the SE of prediction is up to 25% larger than for an average sire. When fewer than 50 sires are used to develop conversion equations, the accuracy of converted EBV of elite sires is very poor, and SE can be 30 to 60% larger than for an average sire. Based on this study, it is recommended that international sire evaluations based on BLUP methodology (rather than linear regression) be made available as soon as possible for nonproduction traits in all countries and for production traits in countries that currently do not participate in routine INTERBULL (International Bull Evaluation Service) analyses.