An approximate formula to predict response to selection for reduced variance of litter size in rabbits is derived under a model postulating that the environmental variance is partly under genetic control. The formula is evaluated using computer simulations. The agreement between predicted and observed results over a period of ten generations of selection is excellent if selection intensity is low to moderate, and less so, as selection increases. An evaluation of the statistical power to detect response to selection for reduced variance under three designs mimicking a prospective experiment involving litter size in rabbits is also reported. Power increases with selection intensity for all three designs and with the duration of the experiment. Divergent selection is the most efficient design followed by a single selected line and then by the design involving a selected and a control line. For divergent selection, power approaches 80% after 6 generations of selection at intermediate and high selection intensities.