By this method I brought my strain within two years up to an average of nearly 90% of the seedlings with a divided primary leaf. Around this average the real numbers fluctuated between the maximum of 99% and the minimum of 70% or thereabouts. This condition was reached by the sixth generation in the year 1894, and has since proved to be the limit, the group of figures remaining practically the same during all the succeeding generations. Such selected plants are very rich in leaves with four, five and six blades. Excluding the small leaves at the tops of the branches, and those on the numerous weaker side-branches, these three groups include the large majority of all the stronger leaves. In summer the range is wider, and besides many trifoliolate leaves the curiously shaped seven-bladed ones are not at all rare. In the fall and in the winter the range of variability is narrowed, and at first sight the plants often seem to bear only quinquefoliolate leaves. [350] I have cultivated a new generation of this race nearly every year since 1894, using always the strictest selection. This has led to a uniform type, but has not been adequate to produce any further improvement. Obviously the extreme limit, under the conditions of climate and soil, has been reached. This extreme type is always dependent upon repeated selection. No constant variety, in the older sense, has been obtained, nor was any indication afforded that such a type might ever be produced. On the contrary, it is manifest that the new form belongs to the group of ever-sporting varieties. It is never quite free from the old atavistic type of the trifoliolate leaves, and invariably, when external conditions become less favorable, this atavistic form is apt to gain dominion over the more refined varietal character. Reversions always occur, both partial and individual. Some instances of these reversions may now be given. They are not of such a striking character as those of the snapdragon. Intermediate steps are always occurring, both in the leaves themselves, and in the percentages of deviating seedlings of the several parent plants. On normal plants of my variety the quinquefoliolate leaves usually compose the majority, when there are no weak lateral branches, or when they are left out of consideration. Next [351] to these come the fours and the sixes, while the trifoliolate and seven-bladed types are nearly equal in number. But out of a lot of plants, grown from seed of the same parent, it is often possible to choose some in which one extreme prevails, and others with a preponderating number of leaves with the other extreme number of leaflets. If seed from these extremes are saved separately, one strain, that with numerous seven-bladed leaves will remain true to the type, but the other will diverge more or less, producing leaves with a varying number of subdivisions.