Plants possess two types of resistance against herbivores: ever-present constitutive resistance and induced resistance triggered by attack. As the production of both resistance types entails a metabolic cost, a tradeoff between them has frequently been hypothesized. Over twenty published studies have tested for the existence of this tradeoff, but this literature is marred by three methodological problems. The first problem is lack of agreement about how to measure induced resistance, a complex trait that typically involves comparison between damaged and undamaged plants. Some metrics of induced resistance confound constitutive and induced resistance, creating evidence for a tradeoff when one does not exist or obscuring real tradeoffs. On both biological and statistical grounds, we argue for the difference in mean resistance between damaged and control plants from the same family or genotype as the best metric of induced resistance. The second problem is that limited sampling (e.g. of families or of individuals within families) or errors in measuring resistance traits of individuals can generate spurious evidence for a tradeoff even when our preferred induced resistance metric is used. The third problem is that some families may show induced susceptibility (lower resistance in damaged than in undamaged plants). To provide a better test for a tradeoff, we devise a Monte Carlo procedure that accounts for sampling variation, measurement error and induced susceptibility without producing unrealistic negative resistance values, and we illustrate it with simulated data. Until the problems we describe are widely addressed and the tools we propose are widely applied, the resistance tradeoff hypothesis cannot be considered to have been adequately evaluated. Our approach also applies whenever the plasticity of a trait (measured as the difference between treatments or environments) is compared to the value of that trait in a single environment.