Constancy is not the result of selection or of improvement. It is a quality of its own. It can neither be constrained by selection if it is absent [563] from the beginning, nor does it need any natural or artificial aid if it is present. Most of my new species have proved constant from the first. Whenever possible, the original mutants have been isolated during the flowering period and artificially self-fertilized. Such plants have always given a uniform progeny, all children exhibiting the type of the parent. No atavism was observed and therefore no selection was needed or even practicable. Briefly considering the different forms, we may state that the full experimental proof has been given for the origin of _gigas_ and _rubrinervis_, for _albida_ and _oblonga_, and even for _nanella_, which is to be considered as of a varietal nature; with _lata_ the decisive experiment is excluded by its unisexuality. _laevifolia_ and _brevistylis_ were found originally in the field, and never appeared in my cultures. No observations were made as to their origin, and seeds have only been sown from later generations. But these have yielded uniform crops, thereby showing that there is no ground for the assumption that these two older varieties might behave otherwise than the more recent derivatives. _Scintillans_ and _elliptica_ constitute exceptions to the rule given. They repeat their character, from pure seed, only in part of the offspring. I have tried to deliver the _scintillans_ from this [564] incompleteness of heredity, but in vain. The succeeding generations, if produced from true representatives of the new type, and with pure fertilization, have repeated the splitting in the same numerical proportions. The instability seems to be here as permanent a quality as the stability in other instances. Even here no selection has been adequate to change the original form. IV. Some of the new strains are evidently elementary species, while others are to be considered as retrograde varieties.