I found that it was not at all necessary to sow any larger quantities of seed, but that the young plants must have room enough to develop into full and free rosettes. Moreover I observed [551] that the attributes of _lata_ and _nanella_, which I now studied in the offspring of my first mutants, were clearly discernible in extreme youth, while those of _rubrinervis_ remained concealed some weeks longer. Hence I concluded that the young plants should be examined from time to time until they proved clearly to be only normal _lamarckiana_. Individuals exhibiting any deviation from the type, or even giving only a slight indication of it, were forthwith taken out of the beds and planted separately, under circumstances as favorable as possible. They were established in pots with well-manured soil and kept under glass, but fully exposed to sunshine. As a rule they grew very fast, and could be planted out early in June. Some of them, of course, proved to have been erroneously taken for mutants, but many exhibited new characters. All in all I had 334 young plants which did not agree with the parental type. As I examined some 14,000 seedlings altogether, the result was estimated at about 2.5%. This proportion is much larger than in the yields of the two first generations and illustrates the value of improved methods. No doubt many good mutations had been overlooked in the earlier observations. As was to be expected, _lata_ and _nanella_ [552] were repeated in this third generation (1895). I was sure to get nearly all of them, without any important exceptions, as I now knew how to detect them at almost any age. In fact, I found many of them; as many as 60 _nanella_ and 73 _lata_, or nearly 5% of each. _Rubrinervis_ also recurred, and was seen in 8 specimens. It was much more rare than the two first-named types. But the most curious fact in that year was the appearance of _oblonga_. No doubt I had often seen it in former years, but had not attached any value to the very slight differences from the type, as they then seemed to me. I knew now that any divergence was to be esteemed as important, and should be isolated for further observation. This showed that among the selected specimens not less than 176, or more than 1% belonged to the _oblonga_ type. This type was at that time quite new to me, and it had to be kept through the winter, to obtain stems and flowers. It proved to be as uniform as its three predecessors, and especially as sharply contrasted with _lamarckiana_. The opportunity for the discovery of any intermediates was as favorable as could be, because the distinguishing marks were hardly beyond doubt at the time of the selection and removal of the young plants. But no connecting links were found.