Five hundred thirty-four steers representing 6 sire breed groups were evaluated in 1994 (n = 282) and 1995 (n = 252) to determine the efficacy of using real-time ultrasound and other live animal measures to predict beef carcass composition. Within 5 days prior to slaughter, steers were ultrasonically measured for 12-13th rib fat (UFAT), longissimus muscle area (UREA), rump fat thickness (URPFAT), and body wall thickness (UBDWALL). Carcasses were fabricated to determine boneless, totally trimmed retail product weight (KGRPRD) and percentage (PRPRD). Correlation coefficients between UFAT and UREA with carcass 12-13th rib fat (CFAT) and carcass longissimus muscle area (CREA) were.89 and.86, respectively. Mean differences indicated UFAT was.06 cm less (P <.01) than CFAT, and UREA was.71 cm2 greater (P <.01) than CREA across both years. Carcass measurements were more accurately evaluated with ultrasound in 1994 than in 1995 (P <.01). Regression equations to predict KGRPRD and PRPRD were developed using either live animal or carcass traits as independent variables. Final models (P <.10) using live animal ultrasound variables included live weight (FWT), UFAT, UREA, and URPFAT for KGRPRD (R2 =.84) and UFAT, URPFAT, UREA, UBDWALL, and FWT for PRPRD (R2 =.61). Comparatively, equations using carcass yield grade variables resulted in R2 values of.86 and.65 for KGRPRD and PRPRD, respectively. When equations developed from 1994 steers were applied to steers in 1995, correlations between values predicted from live animal models and actual carcass values were.92 for KGRPRD, and ranged from.73 to.76 for PRPRD. Similar correlations were found for equations developed from carcass measurements (r =.94 for KGRPRD and.80 to.81 for PRPRD). Both live animal and carcass equations overestimated (P <.01) actual KGRPRD and PRPRD. Regression of predicted values on actual values revealed a similar fit for live animal and carcass equations. This research indicates that real-time ultrasound measurements taken on live beef cattle can be used to effectively predict carcass composition.