Purebred Beefmaster steers (n = 160) from five owners were fed at a commercial feedlot in South Texas beginning in November of 2004; 68 steers possessed pedigree information. Beginning in mid-November, steers were individually weighed and evaluated for ultrasound body composition at 56-d intervals by a certified technician. Feeder calf frame (FRM) and muscle (MUS) scores were assigned at initial ultrasound evaluation. Steers were fed and marketed through a lean-based, branded beef program and were harvested in two groups in May and June of 2005 at a commercial beef plant. Analyses were conducted to investigate the ability to predict carcass traits from the different sources of information available on these cattle. Evaluation of carcass traits were investigated using four sets of independent variables referred to as sources A, B, C, or D and ultrasound scan session (1 – 4). An analysis included initial weight at first scan session (IWT), FRM and MUS as independent variables through GLM procedures. B analyses utilized ultrasound measures of the longissimus area, intramuscular fat, fat thickness, rump fat, and gluteus medius depth along with IWT as independent variables. Multiple regression was performed on each carcass trait using IWT and ultrasound traits at each scan session. Mallow’s CP was used to select a model that best described each carcass trait. C analyses (GLM) utilized variables from A and B analyses combined plus ranch. D analyses (GLM) included variables from C analyses plus sire nested within ranch. Respective R-square values (scan 1 – 4) for marbling score were .02, .04, .05, and .10 using A information, .14, .17, .42, and .54, using B information, .35, .35, .47, and .55 using C information, and .56, .59, .65, and .76 using D information. R-square values ranged from .34 to .86 for carcass weight, .11 to .77 for fat thickness, .06 to .82 for ribeye area, and .10 to .81 for yield grade. Ultrasound data obtained closer to harvest and increasing amount of data related to genetic and management background showed increased R-square values, but may be best utilized in conjunction with one another to predict carcass traits and final carcass value.