More than 800 wells in the glacial aquifer system of the Northern United States were sampled for arsenic as part of U.S. Geological Survey National Water-Quality Assessment (NAWQA) studies during 1991-2003. Elevated arsenic concentrations (greater than or equal to 10 micrograms per liter) were detected in 9 percent of samples. Elevated arsenic concentrations were associated with strongly reducing conditions. Of the samples classified as iron reducing or sulfate reducing, arsenic concentrations were elevated in 19 percent. Of the methanogenic samples, arsenic concentrations were elevated in 45 percent. In contrast, concentrations of arsenic were elevated in only 1 percent of oxic samples. Arsenic concentrations were also related to ground-water age. Elevated arsenic concentrations were detected in 34 percent of old waters (recharged before 1953) as compared to 4 percent of young waters (recharged since 1953). For samples classified as both old and methanogenic, elevated arsenic concentrations were detected in 62 percent of samples, as compared to 1 percent for samples classified as young and oxic. Arsenic concentrations were also correlated with well depth and concentrations of several chemical constituents, including (1) constituents linked to redox processes and (2) anions or oxyanions that sorb to iron oxides. Observations from the glacial aquifer system are consistent with the idea that the predominant source of arsenic is iron oxides and the predominant mechanism for releasing arsenic to the ground water is reductive desorption or reductive dissolution. Arsenic is also released from iron oxides under oxic conditions, but on a more limited basis and at lower concentrations. Logistic regression was used to investigate the relative significance of redox, ground-water age, depth, and other water-quality constituents as indicators of elevated arsenic concentrations in the glacial aquifer system. The single variable that explained the greatest amount of variation in the data was redox. Multivariate models that included a redox variable overestimated the percentage of samples with elevated arsenic concentrations because, even though elevated arsenic concentrations were associated with strongly reducing samples, not all strongly reducing samples had elevated arsenic concentrations. Arsenic concentrations and redox conditions differed among four broad areas of the glacial aquifer system. For the East, Central, and West-Central north areas, there was a trend of increasing arsenic concentrations that corresponded to an increase in reducing conditions. For the West-Central south area, arsenic concentrations in oxic samples were higher than for the other areas, possibly because of high concentrations of orthophosphate, which is linked to desorption of arsenic from iron oxides under oxic conditions. The observed differences in arsenic concentrations among broad areas of the glacial aquifer system were generally consistent with a conceptual model developed by Smedley and Kinniburg, who studied or reviewed studies of widespread arsenic contamination in Bangladesh, India, China, Vietnam, Hungary, Argentina, northern Chile and the Southwestern United States.