Trade-offs in fitness on different host plants has been a central hypothesis in explaining the evolutionary specialization of herbivores. Surprisingly, only a few studies have documented such trade-offs. In this paper, I present results from a selection experiment that demonstrates trade-offs in host plant use for a polyphagous spider mite. Although adaptation to a novel poor-quality host did not result in detectable costs on a favorable host, spider mites that had adapted to a poor-quality host lost their ability to tolerate the poor-quality host when they were reverted to the favorable host for several generations. Trade-offs in fitness on alternative hosts among herbivorous spider mites remains one of the classic empirical examples of constraints on the evolution of host range. Adaptation to the novel poor-quality host was not associated with adaptation to a related host-plant species or to particular host-plant chemicals that I assayed. Thus, the complexity of host-plant defenses may restrict host shifts to single species of novel host plants, and adaptive zone shifts onto entire groups of plants predicted by the Ehrlich and Raven Model may be rare. Spider mite performance was genetically associated with host-plant preference. Mites from the control population showed a significant preference for the favorable host plant, whereas mites adapted to the novel host plant showed no preference. Finally, although induced plant responses to herbivory in the poor-quality host decreased the fitness of unselected mites, induced responses resulted in higher fitness of adapted mites. These results suggest that spider mites that rapidly adapt to particular host plants can overcome consti- tutive and inducible plant defenses.