A major enigma in insect ecology is why many more host plant specialists than generalists exist, especially since laboratory and field experiments have indicated that specialists are usually able to use a much broader spectrum of plants than is observed in nature. It has been suggested that perceptual constraints coupled with considerations of foraging efficiency, may play a role in the evolution of specialization. Here we investigate this notion in the context of insects selecting different types of plants, where the ratios of particular compounds in a blend of odorants provide the cues necessary to discriminate different plants from one another. The discrimination task is modelled by feedforward neural networks that are identified with particular individuals. These networks replicate with the clonal reproduction of individuals, and the strength of the synapses in these networks are able to mutate from one generation to the next. Individuals exploit the different plant types at a level determined by their perceptual response to the individual plants of each type. The fitness of individuals is determined by the relative nutritional value of each plant type and the proportion of individuals in the population exploiting that plant type. From simulations of the evolution of the perceptual networks, we are able to conclude that the probability of particular preferences evolving depends on how close the signals of the plants of different nutritive values are to one another. At reasonably high mutation rates the more easily implemented plant preferences evolve earlier and are at a competitive advantage compared with later evolving, equally fit plant preferences. At low mutation rates, evolution stalls for long periods of time, but when change does occur it is saltational, Evolutionary equilibria typically involve guilds of complementary species that together constitute an 'ideal free distribution' in terms of the productivity of the different plants Our results also suggest: that the mix of phenotypes in these guilds is critically dependent on the order of appearance of various combinations of specialist and generalist phenotypes; that this order depends on the difficulty of the perceptual task associated with each phenotype; that any differences in the relative utilization by a generalist of different species of plants will lead to the emergence of one or more specialists that exploit the plants most under-utilized by the generalist; and that evolutionary changes in guild structure are less frequent than mutational rates might suggest, but are saltatory when they occur. Consequently, the strategy to specialize may dominate for two reasons: specialization appears to evolve more readily in complex environments; and the ideal free distribution mentioned above is more easily matched by a group of specialists or by generalists in concert with specialists than by a generalist alone. Finally, our analysis suggests the hypothesis that oligophages or heterophages will not omit nutritive plants from their diet that have chemical signatures intermediate between plants upon which these herbivores feed.