Experimental studies suggest a myriad of mechanisms by which inorganic arsenic can interfere with central nervous system development, and, indeed, epidemiological studies published in the last dozen years suggest that exposure to arsenic impairs children's cognitive development. Most of the studies have been conducted in developing countries (e.g., Bangladesh, India, Mexico), where exposure to arsenic is thought to be considerably higher than it is in developed countries. This review summarizes the results of these studies, focusing in particular on issues pertinent to risk assessment, including the existence of critical windows of vulnerability, characteristics of the dose-effect relationships (e.g., the lowest adverse effect level, the functional form), the most sensitive neurodevelopmental endpoints, and potential effect modifiers such as host characteristics (e.g., methylation efficiency, sex) and co-exposures to other neurotoxicants (e.g., lead, manganese). At present, the epidemiological data do not permit firm conclusions to be drawn regarding these issues. Several factors that complicate an effort to compare the results of studies are identified, including use of a variety of indices of external and internal exposure, and inconsistency in the measurement of important potential confounders for neurodevelopmental outcomes.