Computer models can be powerful tools for addressing many problems in fishery management, but uncertainty about how to apply models and how they should perform can lead to a cautious approach to modeling. Within this approach, we expect models to make quantitative predictions but only after all model inputs have been estimated from empirical data and after the model has been tested for agreement with an independent data set. I review the limitations to this approach and show how models can be more useful as tools for organizing data and concepts, learning about the system to be managed, and exploring management options. Fishery management requires deciding what actions to pursue to meet management objectives. Models do not make decisions for us but can provide valuable input to the decision-making process. When empirical data are lacking, preliminary modeling with parameters derived from other sources can help determine priorities for data collection. When evaluating models for management applications, we should attempt to define the conditions under which the model is a useful, analytical tool (its domain of applicability) and should focus on the decisions made using modeling results, rather than on quantitative model predictions. I describe an example of modeling used as a learning tool for the yellow perch Perca flavescens fishery in Green Bay, Lake Michigan.