[Illustration: FIG. 87. Cross of normal and fantail pigeons. (See Fig. 86.) The F_2 range is wider than that of F_1. The normal grand-parental type of 12 feathers was recovered in F_2 but the higher numbers characteristic of fantails were not recovered.] The preceding account attempts to point out how I should prefer to interpret the problem of selection in the light of the most recent work on breeding. But I would give a very incomplete account of the whole situation if I neglected to include some important work which has led some of my fellow-workers to a very different conclusion. [Illustration: FIG. 88. Scheme to show classes of hooded rats used by Castle. (After Castle.)] Castle in particular is the champion of a view based on his results with hooded rats. Starting with individuals which have a narrow black stripe down the back he selected for a narrower stripe in one direction and for a broader stripe in the other. As the diagram shows (fig. 88) Castle has succeeded in producing in one direction a race in which the dorsal stripe has disappeared and in the other direction a race in which the black has extended over the back and sides, leaving only a white mark on the belly. Neither of these extremes occurs, he believes, in the ordinary hooded race of domesticated rats. In other words no matter how many of them came under observation the extreme types of his experiment would not be found. Castle claims that the factor for hoodedness must be a single Mendelian unit, because if hooded rats are crossed to wild gray rats with uniform coat and their offspring are inbred there are produced in F_2 three uniform rats to one hooded rat. Castle advances the hypothesis that factors--by which he means Mendelian factors--may themselves vary in much the same way as do the characters that they stand for. He argues, in so many words, that since we judge a factor by the kind of character it produces, when the character varies the factor that stands for it may have changed. As early as 1903 Cuénot had carried out experiments with spotted mice similar to those of Castle with rats. Cuénot found that spotted crossed to uniform coat color gave in F_2 a ratio of three uniform to one spotted, yet selection of those spotted mice with more white in their coat produced mice in successive generations that had more and more white. Conversely Cuénot showed that selection of those spotted mice that had more color in their coat produced mice with more and more color and less white. Cuénot does not however bring up in this connection the question as to how selection in these spotted mice brings about its results. Without attempting to discuss these results at the length that they deserve let me briefly state why I think Castle's evidence fails to establish his conclusion.