Experimental data on the rate of response to artificial selection in initially inbred lines or the rate of divergence among inbred sublines can be used to estimate the rate of increase in variance of quantitative traits from new mutations. So far estimates have been based on the infinitesimal model of many genes with small additive effects which imply a rate of increase in heritability for Drosophila melanogaster bristle number traits of about 0.1% per generation. Such estimates are biased because mutants tend to have large effects, to have non-additive gene action, and to be deleterious. Here, recent information on the distribution of effects of new mutations on Drosophila melanogaster bristle number and viability is used to infer the direction and magnitude of this bias. The infinitesimal model tends to underestimate the mutational variance, typically by a factor of about 3, but this factor depends on the experimental design. Averages of revised estimates, accounting for this bias, of the per generation increment in heritability from mutation are 0.36% and 0.21% for abdominal and sternopleural bristle number, respectively, in experiments involving M strains, and 1.4% and 0.7% for abdominals and sternopleurals, respectively, in P strains.