Theoretical predictions of changes in variance with disruptive selection have used models of infinitely many genes so the increase in variance was necessarily due to linkage disequilibrium. With small numbers of loci, the disequilibrium is shown still to comprise the major part of the changes in variance.In a replicated experiment with Drosophila melanogaster, disruptive selection was practised for three generations, and this was followed by 5 or 7 generations of random mating. The heritability, as estimated from regression of progeny on parent, rose from 37% to 68% on selection, and subsequently declined to 45% on random mating. Changes of variance can be interpreted invoking the build up of linkage disequilibrium during selection followed by its breakdown upon relaxation. The results agree well with those obtained from Monte Carlo simulation.