Some problems in ecology call for models that are challenging by virtue of being large in scope of processes included, computationally expensive, data intensive, or crossing multiple scales of space or time. These models can be structured well or poorly in concept, mathematical formulation, and computational formulation. Thus, the need for high-performance computing may be driven either inherently or by inefficient formulation. Further, developing a model involves many decisions that can lock in future development and so appropriate care is mandated for developers. Scoring "well" on all counts is important for predictive models that are needed in studies of climate change, conservation, biogeochemistry, and other fields. This session offers a panel-led open discussion on making models with clear goals, good formulation (concise and accurate process description, minimal use of adjustable parameters, mathematical efficiency, compliant code), comprehensive documentation in code and users guides, and effective extraction of results in text and graphics. Moreover, many models are destined to be reduced to more empirical, simpler models for decision support systems, formal or informal. This poses a novel challenge for extracting robust patterns and requires insight about such key features as time scales that collapse many phenomena into relatively unimportant transients. Highly informative about model reliability are analyses of output sensitivity to initial states, driving variables, boundary conditions, bio(physical) parameters and, especially, the dreaded but sometimes necessary adjustable parameters. Several case studies will be discussed and participants can expect varied viewpoints among the panelists and discussions. Please see our URL: http://gcconsortium.com/model_wiki/.