Most evolutionary algorithms concerned with a memory of evolution aim at memorizing and reusing the recipes of past successes (e.g. fruitful operators or fruitful mutation directions). The scheme proposed here follows the opp opposite track, and memorizes the past failures of evolution (unfit offspring) through a virtual individual termed the virtual loser. The underlying metaphor is that offspring should attempt to get further away from the virtual loser than their parents. This is done by a new evolution operator, termed flee-mutation, as an alternative to standard recombination and mutation. Special attention is paid to adjusting the flee-mutation step size. Experiments on large sized problems validate this approach and unexpectedly show that a constant flee-mutation step size over a population is desirable.