Risk-based regulatory decisions generally apply a margin of safety meant to guard against underestimation of risk in the face of inter-subject variability and uncertainty. Since these two components often are unknown or only vaguely characterized, the decisions involved usually employ conservative default assumptions concerning the margin of safety, resulting in regulatory limits that may be more (or less) health protective than necessary if variability and uncertainty could be characterized probabilistically. As a result, it remains impossible in most cases to determine the degree of protectiveness inherent in a standard. The debate about maximum contaminant levels (MCLs) of arsenic is an example. At present, we can only get a vague idea that lowering MCLs results in larger margins of safety, but at the expense of greater compliance costs. If the magnitude of this margin of safety is not taken into account, it is possible that an MCL may be established based on a significantly larger margin of safety than is necessary, reasonable or consistent with that applied to other contaminants. Thus an unnecessarily expensive treatment policy may be selected. In this study, a new framework of probabilistic risk-based decision making was developed. A meta-analysis was conducted for arsenic in drinking water by combining several epidemiological studies from various regions (such as Taiwan, US, Argentina, Chile and Finland). Then the results of the meta-analysis were incorporated into the framework to characterize the margin of safety through variability and uncertainty analyses. The final product of this study is a method of probabilistic risk assessment that better deals with variability and uncertainty issues. This risk assessment methodology can help decision-makers make optimal determinations on regulatory limits for a contaminant that adequately protect human health with an ample margin of safety at a more reasonable cost than currently is the case.