Adaptive resource management is a continuous learning process in which current knowledge always leads to further experimentation and discovery. Adaptive management evolves by learning from mistakes. Designing adaptive management strategies involves four tasks. First, the problem must be defined and bounded. There is growing recognition of the need to define and bound problems at the landscape level. Second, existing knowledge must be readily accessible so that errors can be detected and used as a basis for further learning. The current information structure supporting fire management was designed to support the 10 a.m. policy and is inadequate to support current policy. Expert systems and other recent developments in artificial intelligence can provide the necessary means to develop an accessible repository of current knowledge. Third, the inherent uncertainty and risk surrounding possible future outcomes must be displayed. Bayesian decision analysis can be used to deal with uncertainty and risk. Fourth, balanced policies must be designed. These must provide for resource production and protection while creating opportunities to develop better understanding. Signal detection theory and receiver operating characteristic curve analysis provide tools to help design balanced policy. These concepts are illustrated by applying them to the problems surrounding wilderness fire management and the need for long-range fire danger information.