Establishment of natural background levels (NBL) for groundwater is commonly performed to serve as reference when assessing the contamination status of groundwater units. We compare various selection methods to establish NBLs using groundwater quality data forfour hydrogeologically different areas in the highly populated and developed subcatchment Western River Rhine, The Netherlands: selection of old groundwater (before 1945), of tritium-free groundwater (i.e., infiltrated before 1950), and of groundwater having no agricultural contamination by NO3 and SO4. Differences as well as similarities in percentile valuesfor Cl, NH4, and SO 4 concentrations are observed among the selection methods as well as the spatial units, pointing out that selection of the data set is a crucial step in deducing NBLs. The following general points of attention are deduced: (1) reference to composition of recharge water (rain or river infiltrate) is necessary to confirm the statistical outcomes, (2) old analyses are affected by conservation errors after sampling for redox-sensitive solutes and may be obtained by selective sampling, (3) old analyses are the only direct reference for NBLs for groundwater units having only anthropogenically influenced, young groundwater at present and (4) establishment of a priori percentile values as maximum NBL is not right and confirmation by additional process-based insight in the controls on water composition is necessary. © 2008 American Chemical Society.