Divergent artificial selection for abdominal bristle number in Drosophila melanogaster has been carried out starting from a genetically homogeneous base population. Lines with two different systems of mating, random (P lines) or between full sibs whenever possible (about 50%), random otherwise (I lines) were compared. Responses after 40 generations of selection were mostly due to one or two mutations of large effect (0·2 to 2 phenotypic standard deviations) per line. Ten mutations affecting the selected trait were individually studied (five lethal and five non-lethal, these being predominantly additive). These mutations satisfactorily explain the response attained, although some minor mutations may also be involved. No evidence of epistasis for bristle number was found. The average final divergence was 57% larger in the P lines, but it was mostly due to lethals or highly deleterious mutations. Thus, after relaxation of selection, the ranking reversed and the mean divergence became significantly larger in the I lines (14%). Analysis of inbreeding showed that the very small amount of variation created by spontaneous mutations (a heritability for the selected trait of about 3%) was responsible for a reduction in the effective size of about 50% in the I lines (relative to the case with random selection), but only about 10% in the P lines. Mutational heritabilities estimated from the response to selection (0·05–0·18%) were within the range usually found for this trait in previous experiments. REML estimates account for correlations between relatives, and were much larger in those lines where the response was due to lethal mutations, as these do not contribute to response after reaching maximum frequency.