Numerous coffee progenies, particularly of the bourbon variety of Coffea arabica, have been analysed during these last 25 years in the Central Experiment Station, Campinas. It has been observed that the variability of total yield of ripe fruits among the plants of the coffee progenies that have been harvested for 12 to 15 consecutive years, is usually very large. Variability of the same order has also been observed among F1 generation plants obtained by crossing selected coffee trees. Four Si progenies and three F1 generations were selected in order to investigate the main causes of this variation. In each S1 progeny and F1 hybrid, plants were selected with the lowest and highest yields (table 2) and also a plant with yield close to the median class; these were selfed in order to obtain S2 and F2 generations. These S2 and F2 progenies were planted in a comparative trial, where also new S1 and F1 generations of the same coffee trees, as well as progenies of selected coffee plants belonging to the other commercial varieties as "Yellow Bourbon" and "Mundo Novo", were included. The trial comprised 56 treatments planted in 20 randomized blocks, each plot consisting of a single coffee tree. The results of the first four harvests (1955-1958) and of the measurements of tree height and vigor are summarized in table 3. The data indicate that the large variation observed among the members of a Bourbon coffee progeny seems to be due mainly to environmental influences, since no differences were found in the yield of the S2 and F2 progenies derived from the S1 and F1 plants respectively, with the highest, median and lowest yields (table 6). The same condition was found in relation to plant height. The yield variance values for the S1 and S2 progenies in each group of progenies do not indicate any specific effect of selfing. In the F2 generations the variances were more uniform and no significant differences were found among the variances of the F1 and F2 generations. An important general information may be drawn from this experiment: the selection of the bourbon variety must be based on the average yield of the progenies of selected plants. Selection of individuals belonging to the same progeny does not lead to any yield increase in this variety, as judged by the results here discussed. The data also explain why very often a low correlation is found between the yield of the mother trees and of their respective progenies. The comparative study of all treatments of this trial indicated that a fair correlation exists between a good vegetative aspect of the progeny and its yielding capacity. It was also noticed that the "Mundo Novo" coffee is a very high yielding variety confirming the results already available about the economic value of this commercial variety of coffee.