Starting from a genetically variable stock homo- or hemi-zygous for Bar two selection lines were set up, one selected for decreased sensitivity to the effect of larval temperature on eye size (‘canalization line’), the other for increased sensitivity (‘anti-canalization line’). In each generation a sample of larvae was grown at 25°C. throughout life and another sample at 25°C. for the first 48 hours, followed by 18°C. until emergence. In the canalization lines a selection was made of individuals (five males and five females out of 100) least affected by the temperature treatment and the anti-canalization line for a similar number most affected by the treatment. These ten males and females were allowed to mate at random and from the eggs produced random samples were then treated in the next generation in a similar manner in the two temperatures. Precise counts of facet numbers on the right eyes were made at generations 4 and 9 and it was clear that selection had been effective both in decreasing and increasing temperature sensitivity.

The whole canalization line can be regarded as a population which has been subjected to disruptive selection for two different criteria: one (small eyes) being regarded as adaptive to the low-temperature regine or habitat; the other (large eyes) being regarded as adaptive to the high-temperature habitat. A similar type of analysis can be applied to the anti-canalization line in which, however, the selective values of the phenotype are regarded as reversed in value in the two habitats. The experimental procedure employed involved not only random mating between individuals selected according to these two criteria but also random allocation of the offspring of this panmictic population to the two habitats of the next generation. This corresponds to a natural population in which there is disruptive selection exerted by two different habitats but no habitat preferences exhibited by members of the population. Under these circumstances, in our experiments, disruptive selection produced only rather slight increases in phenotypic variance, which were rather larger in the anti-canalization than in the canalization lines. Possible reasons for the difference between this result and those reported by Thoday and Gibson are discussed, and it is suggested that an important reason may be the slowness of the response to selection in the first few generations in our lines.