The working model for interpreting revolutionary interactions has itself evolved over the past 10 yr. The original model included two species, each evolving primarily in response to selective pressures exerted by the other; this model implied a perpetual arms race in which each species developed more specialized adaptations to the properties of the other. With increased understanding of ecological and evolutionary patterns, an alternative model of has been invoked. Diffuse coevolution implies that many species, on the same or different trophic levels, may simultaneously exert selective pressures on one another and be affected by changes in other component members. However, no appropriate metaphor has emerged that succinctly describes diffuse coevolution, and conveys the notion of evolutionary changes rippling out in many directions following changes in a component species. Despite its broad acceptance, the concept of diffuse coevolution has been more successful in drawing attention to the complexity of biotic components in evolutionary interactions than in producing clear predictions of outcomes. The strongest cases for stepwise revolutionary interactions, described in the book Coevolution (Futuyma and Slatkin 1983), are those involving a close physical intimacy between organisms, such as endoparasitic helminths and their vertebrate hosts (Holmes 1983). Many of Price's (1980) examples for specialization and numbers of species of herbivorous insects could be construed similarly. Price, in fact, argues that the most speciose families of herbivorous insects, such as cecidomyiids (gall formers), weevils and agromyzids (both with many endoparasites), and aphids, have the most intimate associations with their hosts and are the most specialized species. Therefore, host plant specialists are clustered in particular herbivore groups. Even though specialization increases the potential for stepwise coevolution, many of the host plants involved are used by diverse arrays of consumers, all of which may exert selective pressures on the plants. Coevolution must be evaluated in the context of the larger community. Effects of any change in plant properties may ripple through the entire community. All herbivores feeding on a plant must cope with all the defrnsive properties of that plant, even if only one herbivore species was the dominant selective agent driving a given change. Competitive relations among the plants may shift if the new defense alters resource allocation, herbivory rates, or the plant's ability to cope with herbivory. And the herbivores' interactions with their own natural enemies can be affected by changes in their own population dynamics or in the plant properties to which those enemies may themselves respond. From this viewpoint, a plant property is truly defensive even if it reduces potential herbivory against species not originally responsible for its selection. While stepwise coevolution requires that genetic lineages of interacting species continue to coexist (Futuyma and Slatkin 1983), in diffuse coevolution it is not necessary that lineages of all the main protagonists remain part of the system. The concept of diffuse coevolution also encompasses selective pressures exerted by other trophic levels, including the herbivores' own natural enemies. Many herbivorous insects have evolved morphological, physiological, or behavioral means of reducing the success of their natural enemies. Several papers argue that parasitoids influence the evolution of feeding specializations among herbivores (e.g., Lawton 1986), and Bernays and Graham (1988) rightly point out that predators might also affect the array of plants used by herbivores. Bernays and Graham (1988) argue, however, that generalized predators may lead herbivores to select plants that the predators are less likely to visit, potentially resulting in herbivore feeding specializations. But this argument does not take into account the behavioral flexibility shown by many generalized predators. And though Bernays and Graham (1988) emphasize the lability in plant use by herbivorous insects over a few generations, their argument implies that generalized predators are even less labile than herbivores. In fact, generalized predators (and parasitoids) typically have very labile searching behavior; though sometimes restricted by habitat type or plant architecture, they may be quick to modify searching even among different plant species if prey densities decline locally (e.g., Arthur 1966, Hassell 1978, Lawton 1986). The abilities of predators to change search patterns also typically is much more rapid than when herbivores shift host plant species. Because of these very different time scales,