distribution for various distributions of the input parameters, we could see the sensitivity of the dose distribution to these input distributions. If it turned out that the input distributions had to be extremely and unrealistically variable before the value 3x rule broke down, this would be at least somewhat encouraging. My initial ideas on how to set up the simulation study are as follows | (1 use Bennett's model to illustrate, but the same general scheme could be used on the other models LLL is using): 1. Restrict attention initially to a population of 36 adult women living on Enjebi under famine conditions, where all coconut and other "natural” foods were obtained on Enjebi. That is, choose what LLL currently considers to be the worst possible case, Use the actual diets obtained for the 36 adult females interviewed by Pritchard. For each such female we have an estimate of the grams of food eaten per day for each type of food. We make the assumption that these survey results are representative of the entire population of Enewetak adult females. at this point. There are two options We could use the exact same dietary intakes obtained in the interview, i.e., treat them as constants, or we could assume that the survey result for a given food is the mean of a statistical distribution with some variance (after all, there is certainly error ' involved in estimating daily intakes of food). I prefer the latter approach, but then we must decide what statistical distribution to use and the parameters of that distribution. Initially one might assume a normal distribution with mean equal to the reported survey result and a standard deviation approximated by (b-a)/6, where a and b are imposed lower and upper limits on the distribution (g/day intake for a given food is on the abscissa). Li a b The value of a and b might be initially taken, e.g., to be plus and minus 50 percent of the survey result. With more thought on this, LLL might be able to come up with more likely values for a and b.