1. Strains from a cage population of Drosophila melanogaster were selected for increased body size on the live yeast medium and on two aseptic synthetic media, (1) deficient in protein and (2) with all nutrients reduced to one-third the normal concentration required for growth to normal size. Both these media reduce body size by about 25%.

2. In two strains, mass selected on the live yeast medium, the response continued fairly steadily for at least sixteen generations, when the experiment was discontinued. By this time body size had been increased by some 30%.

3. On the sub-optimal diets the number of selected parents per generation was twice as great as on the live yeast medium, for technical reasons, but the response ceased abruptly after seven or eight generations of selection at a level considerably below that attained by the strains selected on the more favourable diet. Also, when selection was continued on the live yeast diet, no further progress occurred.

4. Flies selected on the different diets and also unselected flies have been grown on the alternative conditions to see how the deviation from unselected is affected. For both strains selected on sub-optimal diets, the deviation from unselected is appreciably greater on the medium used for selection than on the live yeast medium.

5. The response to selection for larger body size on deficient diets can be attributed partly to better adaptation to these conditions. This inference is supported by several lines of evidence. The within-culture variance, which is clearly greater when selected flies are grown on deficient diets, declines with effective selection. Also in the low-protein strain, for which data are available, the duration of the larval period is shortened in the early stages of selection, while egg production considerably exceeds that of unselected flies grown on the same diet.

6. Other effects, of the kind normally selected for on the live yeast medium, also contribute to the variation and selection response on the deficient diets. At first they appear to act more or less independently of the genetic changes which favour increased size via improved adaptation to the diet, but continued selection soon leads to mutual incompatibility between the alternative pathways in growth. Since no further progress occurred when selection was continued on the live yeast medium, the earlier selection had probably lowered the level of adaptation to the live yeast medium. A new genetic situation had been created in which it was impossible to gauge the amount of further progress by reference to the behaviour of strains selected on the live yeast medium from the beginning.

7. Estimates of heritability, based on cumulated selection differentials, are rather similar in the different diets and range between 0·30 and 0·38. On the live yeast medium, the estimate provides a fair guide to future progress, whereas, on the deficient diets, the predictive value is nil since response ceases immediately after the generations which provide the data for the estimates.

8. By comparing the deviation from unselected on the media used for selection and also the other media, alternative estimates of genetic correlation in performance in different conditions can be computed. The estimates were sufficiently divergent to cast doubt on the practical utility of the statistical procedure, which takes no account of the likelihood that individual variation in body size in different environment represents to greater or lesser degree the effects of segregation on different processes of growth and metabolism.

9. Since the course of selection is influenced by nutritional conditions, comparisons of response to selection for the same ‘character’ such as body size, in populations or species adapted to different conditions, must allow for the likelihood that unequal differences between the conditions in which selection is carried out and those in which the animal normally lives may be an important cause of differences in response.