It is a common strategy in surface water quality modeling to attempt to remedy predictive inadequacies by incorporating additional mechanistic detail into the model. This approach reflects the reasonable belief that enhanced scientific understanding of basic processes can be used to improve predictive modeling. However, nature is complex, and even the most detailed simulation model is extremely simple in comparison. At some point, additional detail exceeds our ability to simulate and predict with reasonable error levels. In those situations, an attractive alternative may be to express the complex behavior probabilistically, as in statistical mechanics, for example. This viewpoint is the basis for consideration of Bayesian probability networks for surface water quality assessment and prediction. To begin this examination of Bayes nets, some simple water quality examples are used for the illustration of basic ideas. This is followed by discussion of a set of proposed probability network models for the eutrophication of the Neuse River estuary in North Carolina. The presentation concludes with consideration of applications and opportunities for Bayes nets in predictive water quality modeling.