Abstract.-The phenotypes of gray treefrog (Hyla chrysoscelis) tadpoles vary depending on whether predators are present in the pond. Tadpoles reared in ponds with predatory dragonfly larvae are relatively inactive compared with tadpoles in predator-free ponds, and have relatively large, brightly colored tailfins with dark spots along the margins. Models for the evolution of plasticity predict that induced phenotypes such as this should confer high fitness relative to the typical phenotype when in the presence of predators, but should be costly when the predator is absent. Our study tested for the predicted fitness trade-off in H. chrysoscelis by first rearing tadpoles in mesocosms under conditions that induce the alternate phenotypes, and then comparing the performance of both phenotypes in both environments. We generated the two phenotypes by rearing tadpoles in 600-liter outdoor artificial ponds that contained either two caged dragonflies (Anax junius) or an empty cage. Tadpoles from the two environments showed significantly different behavior, tail shape, and tail color within two weeks of exposure. We compared the growth and survival of both phenotypes over four weeks in ponds where there was no actual risk of predation. Under these conditions, both phenotypes grew at the same rate, but the predator-induced phenotype had significantly lower survival than the typical phenotype, indicating that induced tadpoles suffered greater mortality from causes other than odonate predation. We tested the susceptibility of both phenotypes to predation by exposing them to dragonflies in 24-h predation trials. The predator-induced phenotype showed a significant survival advantage in these trials. These results confirm that the predator-induced phenotype in H. chrysoscelis larvae is associated with fitness costs and benefits that explain why the defensive phenotype is induced rather than constitutive.Key words.-Amphibian, anti predator behavior, Hyla chrysoscelis, inducible defense, morphometry, phenotypic plasticity, polyphenism, predation, trade-off.Received October 17, 1994. Accepted May 9, 1995 Much research on inducible defenses against predators and herbivores is directed toward measuring the fitness of the defensive phenotype, relative to the typical phenotype, in the presence and absence of predation (Kreuger and Dodson 1981;Lively 1986a;Appleton and Palmer 1988;Karban 1993). These costs and benefits are critical because they prove that the phenotype is defensive, and they help explain why the phenotype is inducible rather than constitutive (always present). Models predict that inducible defenses should evolve only when the defensive phenotype is beneficial if expressed in the presence of consumers, but costly in their absence (Lively 1986a;Edelstein-Keshet and Rausher 1989;Baldwin et al. 1990;Clark and Harvell 1992). If the phenotype does not serve a defensive function it will not be favored by selection, and if there is no fitness cost to expressing the anti-predator phenotype in the absence of predators then all ind...