Although a central assumption of most plant-defense theories is that resistance is costly, fitness costs have proven difficult to detect in the field. One useful, though labor-intensive, method to detect costs is to quantify stabilizing selection acting on resistance in field populations. Here, we report on an experimental field study of Solanum carolinense in which we employed a quadratic phenotypic-selection analysis on 12 types of resistance (defined operationally as one minus the proportion of tissue damaged) involving nine species of herbivores. For seven types of resistance, we found significant stabilizing selection and intermediate optimum levels, indicating the existence of fitness costs. These costs were greatest for resistance against frugivores, moderate for florivores, and lowest for folivores. In addition, significant correlational selection gradients were found for ten pairs of resistance measures, which we interpret as evidence that the fitness impacts of these pairs of herbivores combined non-additively. When the herbivores fed on the same type of tissue, their impact was synergistic (greater than additive), and when they fed on different tissues, their impact was antagonistic (less than additive). We suggest this may be a general pattern for correlational selection on resistance to different herbivores.