In the past several years, wetlands have become increasingly used for wastewater treatment purposes. As a consequence, nutrient trapping and accumulation, particularly for phosphorus, have emerged as important environmental management issues. To predict the effectiveness of wetlands for phosphorus removal, data from a large cross-sectional study and from a single wetland over time were obtained with the objective of developing a predictive model. From the analysis of these data, it was found that phosphorus trapping is predictable using a simple function of phosphorus input and water loading. Several statistical models of phosphorus trapping were developed and evaluated using the two data sets. Particular emphasis was placed on generalized additive modeling, a graphics-oriented data-driven approach, for model diagnostic and model-building purposes. A nonlinear model was recommended, and suggestions were made for future work. ¿ American Geophysical Union 1994