Frequency-dependent natural selection is potentially important both as a mode by which gene frequencies of populations change and also as a possible explanation of the genetic load paradox. If most alleles are maintained at equilibrium under a frequency-dependent mode such that they are selectively equivalent at equilibrium, then no load is felt by equilibrium populations. Two main bodies of data frequently are cited as evidence of the actual operation of frequency-dependent selection (Ayala and Campbell, 1974; Lewontin, 1973, 1974). One set of studies compares the history of alleles in populations over time with the predictions from classical, constant fitness selection models. The other set, the subject of this discussion, deals primarily with situations when a choice is made between alternative genotypes. Either a predator makes a choice between alternate prey (Clarke, 1962, 1969; Murdoch, 1969; Murdoch and Oaten, 1975) or an individual chooses a mate from competing genotypes (Petit and Ehrman, 1968; Merrell, 1950). Predator choice is of interest in that it may bias genetic composition of the prey population, while mate choice may bias directly the genetic composition of the chooser's own population. The purpose of this discussion is to point out an ambiguity in interpretation of many reports of frequency-dependent choice. Although this ambiguity occurs in many contexts, published data on minority-male mating advantages in Drosophila will be stressed, because they illustrate particularly well the relation between frequency-dependent choice and frequencydependent natural selection. Several related concepts are applied to these studies: Preference indicates the chooser's bias when it selects among alternate types. If the bias changes with the frequency of chosen types, preference is frequency-dependent; otherwise, it is constant. When selections are always unbiased, we say there is no preference. Propensity refers to the probability that a chooser will make a selection in some finite time interval. (A modifier is normally used with propensity to indicate the specific context, as "mating propensity.") Propensities may be frequency-dependent or constant as are preferences, but they also may vary with frequency of the choosers as well as of the types chosen. Changes or differences in propensity need not imply the corresponding changes in preferences, or vice versa; propensity only indicates how many selections may be expected in a data set. Choice refers to a set of selections made by the chooser as a result of its preferences and propensities. When the chooser has no preference and constant propensity, its choices will be proportional to the availability of the types chosen; such choices are random. Schaffer (1968) describes choice where there are no preferences but inconstant propensity; this will also be considered random choice. Because all other cases produce nonrandom choice, the terms random choice and no preference are equivalent. By analogy, choice should be called "frequency-dependent" only if preference is frequency-dependent, and may be called "constant" if the chooser has a constant bias, or preference, for one of the types chosen. The commonest test for frequency-dependent choice is to reject the null hypothesis of no preference by showing that the frequency of selections is not always