This study investigates the accuracy of an automatic image analysis method that was developed for spiral computed tomography scans (SCTS), with the objective of calculating the volume of muscle in the hind leg (HLMVCT) and lumbar region (LRMVCT) in lambs. The first step in the image analysis method was the isolation (segmentation) of the muscle regions in each image of the SCTS, using a new program that was implemented in the Sheep Tomogram Analysis Routines software (STAR). Due to the differences of muscle shape in the regions investigated, the new segmentation program applies different segmentation paths in specific subregions. These were automatically identified by the program based on skeletal landmarks. After the segmentation was completed, the muscles areas were automatically measured by counting the pixels representing muscle in each image; the volumes were calculated by adding the muscle areas of each image multiplied by the depth of the image (inter-slice distance). The accuracy of these measures of muscle volume was evaluated, using regression analysis, by comparing HLMVCT and LRMVCT to the hind leg and lumbar region muscle weights measured after dissection (HLMWD, no. =240, and LRMWD, no. =50, respectively) of Texel (TEX) and Scottish Blackface (SBF) female and male lambs slaughtered in 2003-04. The effects of breed, sex and year on the association (SCTS v. dissection) were evaluated. There was a strong association between HLMVCT and HLMWD ( R2=97·4%), which only increased slightly ( R2=97·7%) when breed was included in the model. This indicates that HLMWD can be estimated directly from HLMVCT with a high degree of accuracy. For the lumbar region, the association was high ( R2=83·0% to 88·8% depending on the model) but lower than in the hind leg, probably because the automatic segmentation isolates only the areas of the longissimus lumborum and multifidi muscles. Breed had a significant effect on the prediction of LRMWD from LRMVCT, as well as sex in the case of the TEX lambs. The results indicated that the predictions of LRMWD from LRMVCT require different equations for very divergent breeds such as TEX and SBF.