Models that capture underlying mechanisms and processes are necessary for reliable extrapolation of laboratory chemical data to field conditions. For validation, these models require a major revision of the conventional model testing paradigm to better recognize the conflict between model users and model developers risk (as T ype I and Type II errors) in statistical testing of model predictions. The predictive reliability of the models must be hypothesized and tested by methods that lead to conclusions of the form the model predictions are within a factor-of-two of reality at least 95% of the time. Once predictive reliability is established, it can be treated as a method error within a probabilistic risk assessment framework. This report, developed under APM 131 (Develop a Probability-Based Methodology for Conducting Regional Aquatic Ecosystem Exposure and Vulnerability Assessments for Pesticides), describes a step-by-step process for establishing the predictive reliability of exposure models.