This comment on a 2002 article by Theodore Bergstrom in the Journal of Economic Perspectives argues that group or multilevel selection in haystack models does not require assortative mating. If the various subgroups into which the population periodically divides are sufficiently small, random variation will assure that some groups have more altruists than others, and these groups will have more surviving offspring than others. Under conditions in which the subgroups are periodically repooled into a larger population, it is possible for altruists to increase over time in the general population even though they are declining (or not increasing) at all times in every subgroup.