The ability of animal model evaluations to predict the genetic potential of a bull from his sire and maternal grandsire was investigated. Theoretical coefficients were derived for different combinations of progeny records on the bull, sire, and maternal grandsire. Coefficients > 0.50 for sires and > 0.25 for maternal grandsires were associated with bulls with few daughters. Ten animal model evaluations of Holsteins, July 1989 to January 1994, were used to estimate coefficients realized in three populations: 1) all AI bulls (n = 6924), 2) current AI bulls (n = 1344), and 3) elite AI bulls (n = 6116). The PTA were analyzed for milk, fat, and protein yields, and for fat and protein percentages. Birth year of sons nested within the birth year of their sire was included as a random effect with a first-order autoregressive process for the regression model used to estimate the realized coefficients for sires and maternal grandsires. After adjustment for the genetic trend for estimates of sires, the correlation coefficient between predicted merit of sons from 2 consecutive yr ranged from 0.34 to 0.87. The PTA of bulls from first-crop evaluation were accurately predicted from PTA of sire and maternal grandsire for yield and percentage traits.