A technique is proposed for calculating a population risk from exposure to a toxicant, based on human data grouped by exposure levels. Of concern is a health effect that is measured on a continuous (nonquantal) scale. In the absence of response levels considered to be adverse, a range of normal, or abnormal, values can be established for the current population. For any change in the exposure levels, the change in the probability of abnormal values of a health effect can be calculated for each of the subgroups and summed to obtain the risk for the population. This requires establishing a relationship between the health effect and exposure, and a measure of the distribution of effects within subgroups. In the case of normally distributed effects, the only additional information needed is an estimate of the standard deviation for each of the subgroups. Then, changes in probabilities of abnormal values can be estimated from changes in exposures, to aid in risk/benefit decisions for product development or regulation. The procedure is illustrated with data presented by Gladen et al. (1988) on the reduction of the Bayley Psychomotor Development Index for 12‐month old infants exposed to polychlorinated biphenyls (PCBs) transplacentally. The additional probability of abnormally low index scores is calculated for infants potentially exposedin uteroto the release of PCBs from silicone in some types of medical devices. If such exposure occurs, this neurotoxic risk is estimated to be quite low.