gave doubtful figures and were likewise eliminated, and only 6 out of a lot of nearly 300 flowering plants reached an average of 21 for all of the flowers. Our summer is a short one, compared with the long and beautiful summer of California, and it was too late to cut off the faded and the open flowers, and await new ones, which might be purely fertilized after the destruction of all minor plants. So I had to gather the seed from flowers, which might have been partially fertilized by the wrong pollen. This however, is not so great a drawback in selection experiments as might be supposed at first sight. The selection of the following year is sure to eliminate the offspring of such impure parentage. [497] A far more important principle is that of the hereditary percentage, already discussed in our lecture on the selection of monstrosities. In our present case it had to be applied only to the six selected plants of 1895. To this end the seeds of each of them were sown separately, the ray-florets of the terminal heads of each of the new generation were counted, and curves and averages were made up for the six groups. Five of them gave proof of still being mixtures and were wholly rejected. The children of the sixth parent, however, formed a group of uniform constitution, all fluctuating around the desired average of 21. All in all the terminal heads of over 1,500 plants have been subjected to the somewhat tedious work of counting their ray-florets. And this not in the laboratory, but in the garden, without cutting them off. Otherwise it would obviously have been impossible to recognize the best plants for preservation. I chose only two plants which in addition recommended themselves by the average number of rays on their secondary heads, sowed their seeds next year separately and compared the numerical constitution of their offspring. Both groups averaged 21 and were distributed very symmetrically around this mean. This result [498] showed that no further selection could be of any avail, and that I had succeeded in purifying the 21-rayed _grandiflorum_ variety. It is from this _grandiflorum_ that I have finally produced my double variety. In the year 1896 I selected from among the above quoted 1,500 plants, 500 with terminal heads bearing 21 or more rays. On these I counted the rays of all the secondary heads about the middle of August (1896) and found that they had, as a rule, retrograded to lower figures. On many thousands of heads only two were found having 22 rays. All others had the average number of 21 or even less. I isolated the individual which bore these two heads, allowed them to be fertilized by insects with the pollen of some of the best plants of the same group, but destroyed the remainder.