Thirty-nine mature cows were divided into three condition groups on the basis of their subcutaneous fat thickness as determined by real-time ultrasound. A representative animal from each group was measured and slaughtered. The remaining cows with each group were stratified evenly into two groups with one group fed to gain weight and the other to lose weight. Several ultrasound and other live measures were taken every 4 wk and two animals per subgroup were randomly slaughtered. Carcass data were collected and one side of each carcass was boned, ground, mixed, and subsampled for fat and protein determination. Four regression equations were generated to predict percentage of fat (FAT), percentage of protein (PROT), total fat (TOTFAT), total protein (TOTPROT), total calories (CAL), CAL per live weight (CAL/WT), yield grade (YG), and marbling (MARB). The first equation used all live measures (SUB), the second equation used only objective live measures (OBJ), the third equation incorporated traditional live measures (EAS), and the fourth equation used only carcass data (CAR). Adjusted R-squares of the most appropriate equation using the SUB, OBJ, EAS, and CAR measurements were .82, .73, .82, and .82 for FAT; .82, .57, .61, and .66 for PROT; .89, .87, .86, and .85 for TOTFAT; .95, .95, .93, and .74 for TOTPROT; .93, .92, .91, and .90 for CAL; .83, .78, .83, and .82 for CAL/WT; .86, .86, .78, and .93 for YG; and .75, .70, .74, and .74 for MARB, respectively. It seems that condition score or ultrasound with other objective live measures is as accurate in predicting cow composition as carcass measures.