A critical challenge faced by sustainability science is to develop strategies to cope with highly uncertain social and ecological dynamics. This article explores the use of the robust control framework toward this end. After briefly outlining the robust control framework, we apply it to the traditional Gordon-Schaefer fishery model to explore fundamental performance-robustness and robustness-vulnerability trade-offs in natural resource management. We find that the classic optimal control policy can be very sensitive to parametric uncertainty. By exploring a large class of alternative strategies, we show that there are no panaceas: even mild robustness properties are difficult to achieve, and increasing robustness to some parameters (e.g., biological parameters) results in decreased robustness with respect to others (e.g., economic parameters). On the basis of this example, we extract some broader themes for better management of resources under uncertainty and for sustainability science in general. Specifically, we focus attention on the importance of a continual learning process and the use of robust control to inform this process.natural resources ͉ resource management ͉ vulnerability ͉ policy design ͉ environmental policy A s the 4-decade-old debate about resource scarcity and continued economic development (1, 2) continues (3-5), the challenges facing humanity are increasing in scale and complexity. Global climate change, depletion of fisheries worldwide (6, 7), and potential pandemics are vivid examples that highlight a key challenge facing sustainability science: decision making under pervasive uncertainty associated with complex social-ecological processes. Although statistical decision theory and dynamic optimization techniques provide guidance in this regard, they typically rely on the capacity to assign probabilities to uncertain events (8, 9). In cases like those mentioned above in which assignment of probabilities is often impractical or impossible, how can policy be developed?One strategy might be to employ a panacea, a set of general principles that work well in a variety of circumstances. In this paper we use techniques from robust control to show that, although panaceas can be tuned to perform well for a range of ''sufficiently similar'' circumstances, their use can be very dangerous. The danger lies in perceiving different circumstances as sufficiently similar when, in fact, they differ in subtle but important ways. Specifically, we show that it is difficult, if not impossible, to devise a panacea for the set of all possible ''similar'' circumstances generated by the Gordon-Schaefer (10, 11) fishery model with a given range of parametric uncertainty. We find, rather, policies robust to uncertainty in one group of parameters are necessarily vulnerable to uncertainty in another group. Our analysis suggests, for example, that it is not possible to tune policies to achieve good performance over a given range of intrinsic growth rates (i.e., a set of different but sufficiently similar circumstances) w...