A major difference in developmental stability has been demonstrated between two populations produced by artificial selection for supernumerary scutellar bristles. The test system involves the substitution of an X-ray induced partial revertant of sc1 for the wild-type allele at the scute locus, enabling direct comparisons to be made of the degree of canalization at the wild-type level of expression of the character. One population is comparable with the unselected Canberra stock in stability, though it differs appreciably in mean bristle number: the other population shows a marked reduction in the level of regulation of bristle number variability. The alleles responsible for the reduced level of canalization are rare in the base population, and are of particular importance in the determination of limits to directional selection. Their effects on developmental stability have been shown to depend on the activity of the allele at the scute locus.