Selection for reproduction at an early or a late age in life was applied to populations of D. melanogaster for 21 to 29 generations, with two experimental treatments of larval density. Populations with high and uncontrolled numbers of competing larvae responded strongly to selection for late-reproduction with the length of adult life increasing by as much as 50 per cent. In this treatment, selection produced true breeding long- and short-lived lines. When populations of developing larvae were held low, however, longevity fluctuated wildly during selection, showing little overall response, as several previous tests of senescence theory have also found. These experiments suggest that life span is either physiologically limited in that environment, or populations are unable to respond because either phenotype/genotype correlations are reversed, or genetic variation is suppressed. The inability of former studies to obtain a response to selection appears to have resulted from the artifactual introduction of strong gene-environment interactions through the use of a competition-free environment.