In experiments that investigate species’ interactions, individuals are often chosen at random to represent their populations. However, this practice can generate misleading results when individuals of different species do not interact at random. We illustrate this effect by examining oviposition preferences of Euphydryas aurinia butterflies from three populations using three different plant genera. We first offered each insect a randomly chosen member of its own host population and a foreign host (Succisa pratensis) not present in the insect’s habitat. The butterflies uniformly preferred the foreign Succisa over their own hosts. Preferences were apparently maladaptive because insects wasted time searching for a nonexistent plant. We repeated the experiment using individual hosts that had naturally received eggs in the field. The overall preference for Succisa and the appearance of maladaptation both disappeared. In the original experiments, our random choice of experimental host individuals had combined with strong within-species discrimination by the butterflies and with overlap of acceptability between host species to obscure the true nature of host preference.