extreme non-linearity of the dose-response relationship being tested. If the relationship were linear no difficulty would arise, because the mean response to the non-uniform irradiation would be identical with the response at the mean dose. Unfortunately, it is not uncommon for attempts to be made to fit non-linear doseresponse curves by relating tumour yields to average dose in circumstances where thedose in the cells at risk is likely to be very non-uniform, for example, the induction of osteosarcoma by isotopes of radium in the skeleton as discussed in reports of the National Academy of Sciences Committee on the Biological Effects of Radiation (BEIR)2 and the United Nations Scientific Committee on the Effects of Atomic Radiation (UNSCEAR).3 To allow for dose heterogeneity in such cases it is necessary to have some estimate of the way in which the dose varies in different parts of the irradiated site. If a fraction f(D)AD of the cells at risk receive doses in the range D to D +AD, so that the mean dose among all cells is D_-D f(D)dD, and if the true dose-response relationship were y = bDe-.D, the mean response amongst the cells at risk due to the dose distribution f(D) would be y-JbDe-XD f(D)dD and not bDe-D as used by Smith and Doll. It is evident that the authors are aware of this problem, for example, from their discussion about the lack of an excess of leukaemia in patients treated for cancer of the cervix in whom parts of the marrow received either very large or very small doses of radiation. Reanalysis of the data to allow for this complication would doubtless require substantial clerical and computing work, and we may have to wait some time for it. We hope, however, that the authors will carry out analyses along the above lines as soon as possible.