A divergent selection experiment for carcass leanness in Texel-Oxford sheep was established to examine the differences between genetically lean and fat animals derived from the same base population. The selection criterion was designed to change body composition without a corresponding change in live weight, using an index of ultrasonic backfat depth and live weight at 20 weeks of age. The index was constructed using estimates of genetic and phenotypic parameters, which were available at the start of the experiment in 1985. The difference between the high and low lines, after 3 years of selection, for the selection index, live weight at 20 weeks of age, ultrasonic backfat and muscle depths was 0·59,1·30 kg, -0·76 mm and 0·88 mm respectively. The estimated difference for carcass lean and fat weight was 0·72 kg and -0·04 kg respectively with 13·5 g/kg and -13·8 g/kg for carcass lean and fat proportion. The realized heritability for the selection index, estimated from the regression of cumulative response on cumulative selection differential was 0·42, with a standard error of 0·25, after taking account of genetic drift. The selection index heritability, estimated with residual maximum likelihood (REML) methodology, was 0·46 (s.e. 0·14). REML heritability estimates for live weight at 20 weeks of age, ultrasonic backfat and muscle depths were 0·20 (s.e. 0·13), 0·35 (s.e. 0·14) and 0·43 (s.e. 0·14) respectively.