A large scatter of the experimental points would not be unexpected under such conditions and in fact the agreement between such very heterogeneous data ap- pears rather surprising. ‘The variability on each experimental point (which is available in many of the experiments, although not in all) has not been plotted since it would be expected to be reabsorbed in the variability between series and could not in any case be used to weigh the points in the analysis to follow. In order to avoid including animals dying from early radiation effects, mice surviving less than 60 days were excluded. The analysis was limited to doses of up to 900 rad and corresponding maximum effects of about 60 per cent. In order to standardize the abscissa dose scale a conversion factor of 1 R= 0.95 rad was used. The ordinate scale is simply the percen- tage of life-span-shortening (calculated from mean or median values as they were available) by comparison with the life-span of non-irradiated mice, irrespective of the duration of life of the normal animals or of the pathology at death. 91. The nature of the plot in Figure II is such that for very high doses a saturation of the effect must become manifest, although it may reasonably be assumed that within 50 - 60 per cent no saturation might distort the plot. In the absence of any information as to the possible form of the dose-effect relationship a non-weighted linear regression was first interpolated to the data, according to the formula y where b y = atodD is the precentage of life-shortening, are the coefficients of the regression. (5) D is the dose and a and The calculated least-square solu- tion to the above equation was y = 1.615 + 0.0502 D and it gave an Ro value of 0.788. (6) Although at inspection of the data a higher-order component was not clearly apparent, its existence could not be excluded and therefore the following relationship was also fitted y = atbD+eD (7)