Logistic regression models and other methods are frequently used to assess the probability of nitrate concentrations above a threshold level and predictions of concentrations.  A logistic regression model has been developed to assess the probability of nitrate above 4 mg/L and a non-linear regression model been developed to predict nitrate concentrations in networks of domestic wells in the glacial aquifer system.  The probability model is based on explanatory variables within a 500m buffer of a well; whereas, the prediction model is based on variables averaged across much larger areas, generally several square miles or more (network averages).  These two different approaches for assessing nitrate concentrations in domestic wells resulted in well depth, recharge, soil type, casing diameter, and fertilizer application to be the most important explanatory variables in  both models.  In addition, because the probability model was based on data at individual wells, ground-water age improved the fit of that model.  The prediction model was based on network averages of well data, so general information on type of surficial material improved the fit of the prediction model.  Variables in probability or prediction models can be grouped into indicators of nitrate source, aquifer susceptibility, and geochemical conditions that can affect nitrate concentrations.  There is overlap in these groups, especially in regard to geochemical conditions.