Informed debate on agricultural nonpoint pollution requires evaluation of regional water quality in relation to management practices. It is prohibitive, in terms of cost and time, to run the site-specific process models for regional policy analysis. Therefore, a simplified and robust technique--metamodeling--is suggested to evaluate regional water quality. Data from an experimentally designed simulation of complex surface water and groundwater process models, PRZM and STREAM, are used to develop statistically validated metamodels. The estimated metamodels were integrated with a regional agricultural economic decision making model to evaluate the surface water and groundwater loadings of 16 major corn and sorghum herbicides. Spatial probability distributions are derived for herbicide concentrations exceeding the toxicity-weighted benchmark from the EPA. We estimate that 1.2 percent of the regional soils will lead to groundwater detection of atrazine exceeding 0.12 ?/L, which compares well with the findings of the EPA's groundwater monitoring survey. We find no-till practices to significantly reduce the surface water concentration of atrazine and other herbicides with less impact on groundwater contamination suggesting indirect gains to soil conservation policies. But we also note that an atrazine ban could lead to increased soil erosion, even with the conservation compliance provisions fully incorporated.