Organisms can often benefit by distinguishing between different classes of individuals. An example is kin recognition, whereby individuals preferentially associate with or aid genetic relatives that bear matching recognition cues but reject others. Despite its potential benefits, however, kin recognition using genetically based cues is often weak or absent [1-4]. A general explanation, termed "Crozier's effect," is that when individuals interact randomly, rarer cue alleles less often match cues of other individuals, and so are involved predominantly in "reject"-type interactions. If such interactions are more costly, positive frequency-dependent selection will erode the cue diversity upon which discrimination depends [4, 5]. Although widely cited [1, 2, 4, 6-9], this idea lacks rigorous testing in the field. Here, we show how Crozier's effect applies to interactions between hosts and conspecific parasites, and measure it using field data. In the wasp we studied, conspecific parasitism fits a key assumption of Crozier's model: the same females act as both hosts and parasites. By exchanging offspring between nests experimentally, we find no evidence that females respond to genetically based cues associated with foreign offspring. Through measuring costs and benefits, however, we demonstrate a strong Crozier effect: because more parental investment is wasted when foreign offspring are rejected, interactions involving rejection have substantially lower payoffs than interactions involving acceptance. Costly rejection can thus eliminate cue diversity by causing selection against rare cue alleles, consistent with the absence of genetically based recognition that we observe. Females instead appear to rely on non-genetic cues that enable them to detect less than half of parasitic offspring.