We would like to thank Drs. Richardson and Cole (Richardson and Cole 2011) for taking an interest in a recent paper on multi-model inference (MMI) based on the Japanese A-bomb data for leukaemia mortality from members of our group of research collaborators (Walsh and Kaiser 2011). We also appreciate the time that they took to consider our methodology. This methodology has been successfully applied in many other fields of researchphysics, biology, environmental science, etc.,-where one can find a large number of papers (e.g. Liddle et al. 2006;Zhang and Townsend 2009;Lavoué and Droz 2009). However, we consider the application to be new in the field of radiation epidemiology and are particularly interested in refining and improving our initial approach. We prefer to use the MMI terminology rather than ''model averaging'' since model averaging implies that the related uncertainties from model combinations are reduced-in reality with MMI, the uncertainties are increased to account for uncertainties between several models that describe the data almost equally well.Richardson and Cole correctly state in their comments that ''model averaging is one approach to characterizing uncertainty in risk estimates in epidemiological studies in which there is low statistical power to discriminate between alternative model forms.'' Therefore, we are very surprised and concerned that they caution against our approach in general and state that both of our recommendations concerning the choice of models for risk assessment should be viewed cautiously. Their arguments are based on two examples: one using two models from the appendix of Richardson et al. (2009) and another offering a simply constructed hypothetical numerical illustration of how our approach can lead to biased results. We will consider these two points below and also a third point that illustrates how our initial approach can be improved.Example 1: using two models in the appendix A2 of Richardson et al. (2009) Richardson et al. (2009 specifically addressed the distinction between a modelling approach that they consider to minimize bias in estimation of an association and an approach that focuses on overall goodness of fit. In their analysis of the A-bomb leukaemia mortality data, they adjusted for proximal versus distal location while previous analyses, which did not seem to be cited, had not. Their adjustment was achieved with a four-level variable in the baseline function, which indicated proximal compared to distal location at the time of bombing in each city. The present authors have reproduced the fit of this model and found that the two location parameters for Nagasaki were not statistically significant (p-value = 0.23, 0.15).Richardson and Cole noted that ''prior research suggested that location was a potential confounder of the radiation dose-leukemia association, because rural (i.e. distal) location was a determinant of estimated DS02 dose and rural cohort members may have different mortality risks than urban.'' They state that ''adjustment for this variable di...