1. Mutant stocks whose phenotype is affected by temperature were kept at 18° and 25° C., and selection was applied in an attempt to reduce the magnitude of the phenotypic difference caused by the temperatures. The stocks used were Bar, dumpy, cubitus-interruptus and aristopedia-Bridges.

2. When the stocks were kept for one or more generations at one temperature, then transferred for the next generation or two to the other, and so alternately, selection applied against the phenotypic effect of the temperature was only slightly effective in reducing the differences between the high- and low-temperature phenotypes. It is suggested that this was due mainly to the fact that such selection would be expected to operate on quantitatively-acting genes as well as on genes controlling developmental buffering.

3. With family selection, in which the offspring of a pair-mating was divided into two lots, one kept at each temperature, and selection was made on the basis of the differences in family means, progress was rapid. In the unselected Bar stock, the facet number at 18° was initially about three times that at 25°; after six generations of family selection the difference had been reduced to about 10%, the phenotypes at both temperatures being about intermediate between those seen at the beginning of the experiment. With aristopedia the reduction in the difference in family means was also striking, but in this case was achieved by increasing the abnormality of the 25° phenotype, and hardly at all by lowering that of the 18° population. It is pointed out that this is probably the result of an inadequate scoring system.

4. Although the selection in Bar stocks was effective in increasing developmental canalization against external environmental changes (temperature), it had little or no effect on the asymmetry of the facet numbers.