This paper presents a conceptual framework for identifying the bias that results when aggregate (county-level) data are used to analyze the impact of agricultural activities on groundwater contamination. The use of aggregate data is often necessary due to dimensionality problems or data limitations in national or even regional analysis. With aggregate data, detailed site characteristics are not considered. It is shown that ignoring detailed site characteristics leads to five potential sources of bias. The factors that determine the signs and magnitudes of these biases are examined. Through an application to the state of Wisconsin, three of the five sources of the bias are quantified. The implications of the results regarding the use of aggregate analyses in the design of groundwater protection policies are also discussed. ¿ American Geophysical Union 1995