Eight generations of selection towards a higher longevity were made in a wild strain of Drosophila melanogaster. Two control lines were also observed. Absolutely no response to selection was obtained whilst a major increase in longevity occurred between F2 and F4 in the three lines under observation. It is shown that the major increase in longevity is due neither to genetic drift, nor to changes in classical environmental conditions. The absence of response to selection is demonstrated to be due neither to a too low selection differential, nor to the absence of genetic variability in the strain, nor to inaccuracy in the measurements, nor to recurrent reproduction at an old age. The impossibility to select towards a higher longevity and the total absence of relation between parental and offspring longevities demonstrate that the very large phenotypic variability displayed by longevity in wild strains of D. melanogaster does not depend on a precise set of specific genes or polygenes with additive action. The results are briefly discussed in relation with inbreeding depression and heterosis for longevity and with similar results obtained in experiments of selection for duration of development.