Fig trees and their pollinators, fig wasps, present a powerful model system for studying mutualism stability: both partners depend on each other for reproduction, cooperation levels can be manipulated, and the resulting field-based fitness quantified. Previous work has shown that fig trees can severely reduce the fitness of wasps that do not pollinate by aborting unpollinated figs or reducing the number and size of wasp offspring. Here we evaluated four hypotheses regarding the mechanism of sanctions in four Panamanian fig species. METHODS: We examined wasp and fig samples from field experiments with manipulated levels of pollination. KEY RESULTS: In unpollinated figs, the fig wall and the wasp offspring had a lower dry mass. Unpollinated figs had as many initiated wasp galls as pollinated figs but fewer galls that successfully produced live wasp offspring. Across three experimentally increasing levels of pollination, we found nonlinear increases in fig wall mass, the proportion of wasp galls that develop, and wasp mass. CONCLUSIONS: Our data did not support the hypotheses that lack of pollination prevents gall formation or that fertilized endosperm is required for wasp development. While our data are potentially consistent with the hypothesis that trees produce a wasp-specific toxin in response to lack of pollination, we found the hypothesis that sanctions are a consequence of trees allocating more resources to better-pollinated figs more parsimonious with the aggregate data. Our findings are completely analogous to the selective resource allocation to more beneficial tissues documented in other mutualistic systems.