Forty-nine bulls, 27 Limousin × Friesian and 22 Charolais × Friesian, were evaluated and slaughtered in four batches of about equal size over 4 weeks. Each batch was of one breed. Age, live weight at evaluation and subjective assessments of fatness and conformation were recorded together with fat and muscle measurements by the Delphi, Meritronics, Scanogram, Vetscan, Kaijo Denki, Warren and the Velocity of Sound ultrasonic machines. Experienced operators were used to assess the performance of machine/operator combinations likely to be achieved in bull performance testing in the field. Fat thicknesses and areas, and m. longissimus areas were taken at the 10th rib and 13th rib, and 3rd lumbar regions by most machines. For the Delphi and Meritronics machines, fat thicknesses only were taken; for the Velocity of Sound machine, time interval measurements and anatomical distances were taken at the shoulder, mid back, lumbar and hind limb regions. The left side of each carcass was fully separated into lean, subcutaneous fat, intermuscular fat, bone and waste. All measurements were examined as potential predictors of carcass composition in step-wise regression in a model which included week of evaluation, breed and live weight at evaluation as the first independent variable. On the whole, scanning machines had a higher precision than A-mode machines, with the Velocity of Sound machine achieving the highest precision for carcass lean (g/kg) (residual s.d. = 13·0) and fat (g/kg) (residual s.d. = 14·1). None of the linear and area measurements taken on the carcass achieved the degree of precision of the Velocity of Sound, Scanogram and Vetscan machines.