Abstract. Can the evolution of plant defense lead to an optimal primary production? In a general theoretical model, Loreau (1995) and de Mazancourt et al. (1998de Mazancourt et al. ( , 1999 have shown that herbivory could increase primary production up to a moderate rate of grazing intensity through recycling of a limiting nutrient, provided several conditions are fulfilled. In the present paper, we assume: (1) grazing intensity is controlled by plants through their level of palatability; and (2) plant fitness is determined by its productivity. We explore the conditions under which such an optimal production may be reached through natural selection. We model two competing plant types that differ only in palatability and are distributed in a patchy landscape determined by the plant-herbivore interaction. Patch size is determined by herbivore behavior: herbivores recycle nutrient homogeneously within patches, but recycle nutrient proportionally to consumption between patches. The model shows that a strategy of intermediate palatability can be adaptive in response to a small herbivore that lives on and recycles nutrient around one or a few individual plants. For moderately small herbivores, plant palatability may evolve towards one of two local convergent strategies, depending on the initial conditions. For medium-to large-sized herbivores, the nonpalatable strategy is always selected. We discuss the functional and evolutionary implications of these results, and suggest that the traditional dichotomy describing antagonistic and mutualistic interactions may be misleading.Key words. Antagonism, grazing optimization, group selection, indirect effect, maximum power principle, mutualism, patchy environment, plant defense, plant palatability, primary production, spatial heterogeneity.Received August 5, 1998. Accepted July 12, 1999 Can plant-herbivore interaction be mutualistic? Plant-herbivore interactions have traditionally been considered antagonistic because herbivores have a negative direct effect on plants through biomass consumption. This assumption has been challenged by the grazing optimization hypothesis, which states that herbivores can enhance plant primary production (McNaughton 1979;Hilbert et al. 1981;Dyer et al. 1986): primary production can increase with low grazing intensity and reach an optimum at intermediate grazing, before production decreases again when grazing intensity becomes too high. Several authors have gone further and have suggested that grazing optimization can lead to mutualistic interactions between plants and their herbivores (Owen andWiegert 1981, 1982;Petelle 1982;Vail 1992), based on a few studies that show a positive effect of herbivory on plant fitness (Paige and Whitham 1987; but see Bergelson and Crawley 1992;Paige 1992;Gronemeyer et al. 1997;Lennartsson et al. 1997Lennartsson et al. , 1998. Such a claim has attracted a lot of criticism (Silvertown 1982;Belsky et al. 1993;Mathews 1994). According to Belsky et al. (1993), no plausible explanation of a real benefit for the individu...