Various approaches have been used to classify large geographical areas into smaller regions of similar water quality or extrapolate water-quality data from a few streams to other unmonitored streams. A combination of some of the strengths of existing techniques is used to develop a new approach for these purposes. In this new approach, referred to here as SPARTA (SPAtial Regression-Tree Analysis), environmental characteristics for each monitored stream are first quantified using a Geographic Information System (GIS) and then regression-tree analysis is used to determine which characteristics are most statistically important in describing the distribution of a specific water-quality constituent. GIS coverages of only the most statistically significant environmental characteristics are then used to subdivide the area of interest into relatively homogeneous environmental water-quality zones. Results from the regression-tree analysis not only define the most important environmental characteristics, but also describe how to subdivide the coverage of the specific characteristic (for example, areas with <26% or ???26% soil clay content). The resulting regionalization scheme is customized for each water-quality constituent based on the environmental characteristics most statistically related to that constituent. SPARTA was used to delineate areas of similar phosphorus, nitrogen, and sediment concentrations (by including land-use characteristics) and areas of similar potential water quality (by excluding land-use characteristics). The SPARTA approach reduced the variability in water-quality concentrations (phosphorus, total nitrogen, Kjeldahl nitrogen, and suspended sediment) within similarly classified zones from that obtained using the US Environmental Protection Agency's nutrient ecoregions.