In two trials, male Suffolk sheep were weaned at birth and reared artificially. Animals were ranked on weight at about 90 days and groups comprising the fastest- and slowest-growing animals were selected. The selected rams were subsequently progeny-tested to the same age, with some progeny reared with their dams in normal field conditions and others reared artificially. In Trial I, performance tests were obtained on 35 rams and progeny tests of seven of these (three low and four high), with records on 34 artificially reared and 144 naturally reared progeny on one farm. The regressions of progeny on sire 90-day weight were 0·29 ± 0·12 and 0·19±0·09, respectively. In Trial II, performance tests were obtained on 86 rams, sampled from five source flocks. Rams were selected in pairs of high and low performance from each source, and both members of the pair sent to the same one of seven farms for progeny testing. Ten pairs had progeny tests, comprising 62 artificially reared and 567 naturally reared progeny. The regressions of progeny on sire weight were, respectively, 0·05 ± 0·06 and 0·13 ± 0·03 if source of sire was fitted in the model, and little different, 0·06 ±0·05 and 0·14 ± 0·03 if source of sire was ignored. Progeny from a 2nd year from some sires in Trial II gave rather smaller regression coefficients: 0·06 or 0·09, with source of sire fitted or ignored, respectively. Taking the two main trials together the pooled estimates (ignoring source of sire) of doubled progeny-on-sire regression for 90-day weight were 0·20 ± 0·10 for artificial rearing (a heritability estimate) and 0·30 ± 0·06 for natural rearing. These results suggest that early weaning is an effective method of selecting sires for growth rate.