Fisher's theorem of natural selection implies that the population genetic variance of quasi-neutral traits should be mostly additive. In the case of fitness component traits, however, that variance would be characterised by a substantial contribution from non-additive loci. In parallel, Robertson's theorem states that selection will change the population mean of a trait proportionally to the magnitude of the genetic correlation between that trait and fitness, which should be weak for quasi-neutral traits or strong for the mean fitness components. Drosophila data from inbreeding and artificial selection experiments are discussed within that theoretical framework. In addition, the process of regeneration by mutation of the genetic variance of a quasi-neutral trait (abdominal bristle number) in a Drosophila population initially homozygous at all loci has been analysed. After 485 generations of mutation accumulation, the levels of additive variance found in this population closely approached those commonly observed in laboratory populations. Furthermore, these values, together with previously reported estimates for natural populations, could be jointly explained by a model assuming weak causal stabilising selection.