Hundred and nineteen lambs from a single abattoir in Norway were scanned using computer tomography (CT) at different equidistances (40, 80, 160 and 320 mm) to perform a virtual dissection of the carcasses, separating muscle, fat and bone tissue. The population of sheep used covered the commercial range of breeds and body composition in Norway and the full range of fat and conformation scores. After CT scanning, the carcasses were manually dissected by trained butchers. The volume and weight of all carcasses were estimated using Cavalieri estimation of the different equidistances of CT slices. The precision and reliability of virtual dissection were estimated from repeated measurements using splitting of carcasses into two halves. The results showed that virtual dissection (r > 0.95) was more precise and reliable than manual dissection (r > 0.80), both for carcass tissue weights and proportions. The correlation between virtual and commercial dissection was high for carcass weight and muscle tissue weight, however, lower for fat and bone tissues. The precision and reliability of virtual dissection, and the correlation between virtual and manual dissection were highest using low equidistance CT scanning (40 mm). There were some biases between virtual and manual dissection, especially for bone tissue. The source of bias can be explained by inaccurate manual dissection performed by the butchers and underestimation of bone using Cavalieri estimation. (C) 2008 Elsevier Ltd. All rights reserved.