Genotoxic substances are generally not permitted for deliberate use in food production. However, an appreciable number of known or suspected genotoxic substances occur unavoidably in food, e.g. from natural occurrence, environmental contamination and generation during cooking and processing. Over the past decade a margin of exposure (MOE) approach has increasingly been used in assessing the exposure to substances in food that are genotoxic and carcinogenic. The MOE is defined as a reference point on the dose-response curve (e.g. a benchmark dose lower confidences limit derived from a rodent carcinogenicity study) divided by the estimated human intake. A small MOE indicates a higher concern than a very large MOE. Whilst the MOE cannot be directly equated to risk, it supports prioritisation of substances for further research or for possible regulatory action, and provides a basis for communicating to the public. So far, the MOE approach has been confined to substances for which carcinogenicity data are available. In the absence of carcinogenicity data, evidence of genotoxicity is used only in hazard identification. The challenge to the genetic toxicology community is to develop approaches for characterising risk to human health based on data from genotoxicity studies. In order to achieve wide acceptance, it would be important to further address the issues that have been discussed in the context of dose-response modelling of carcinogenicity data in order to assign levels of concern to particular MOE values, and also whether it is possible to make generic conclusions on how potency in genotoxicity assays relates to carcinogenic potency.