Different approaches were evaluated to improve the accuracy of carcass yield predictions of Canadian lamb carcasses using manually obtained measurements and dual-energy X-ray absorptiometry (DEXA). Several linear carcass measurements were obtained from a population of commercial lamb carcasses representative of the variability in Canadian slaughter plants (n = 155). Carcass measures were categorized into four sets according to when each measure could be obtained in the slaughter process. Each set of carcass measurements were subjected to stepwise regression and used to develop models for the estimation of lean meat and saleable yield percentages. Tissue depth measures (at the GR site) explained 44% of variation in lean meat yield in hot carcasses and 53% in cold carcasses. When additional parameters were included with cold GR, the regression model explained 61.9% of the variability in lean meat yield. Saleable yield predictions were less accurate (R² < 0.40); the greatest degree of variability was predicted when the model included ribeye area (R² = 0.39). The DEXA scans obtained on carcass sides were able to predict about 78% of variability in carcass lean meat yield and 91% of fat content. This information could be used by the lamb meat industry to establish new carcass classification systems based on more accurate lean meat yield values.