The presence of dormant buds allows grazed plants to compensate for destroyed active meristems. In this article we present a theoretical analysis of the adaptive significance of bud dormancy when the risk of herbivory varies from year to year. Under constant herbivore pressure, selection tends to favor either plants that have no dormant buds and hence no capacity of compensatory growth due to low risk of herbivory, or those that leave most of their buds in dormancy because of high risk of herbivory. We show that when the risk of herbivory varies from year to year, selection will favor intermediate phenotypes having both dormant and active meristems. The intermediate phenotypes represent bet-hedging strategies that have lower vari- ance of seed production than either of the extreme strategies. We also show that intensive herbivory can favor meristem allocation strategies that allow plants to overcompensate for herbivore damage. Our results suggest that two kinds of herbivores might cause selection pres- sures favoring compensatory growth: large ungulate herbivores that remove a large proportion of the aboveground parts of the attacked plants and that predictably attack fairly many plants (>50%) each year, and invertebrate herbivores that periodically cause extensive damage during years of mass attacks.