Dr. Don Hendricks ~-10- August 28, 1972 trials are needed if this is essential. We are not very clear as to how to handle the problem of allocating soil profiles. In "fallout" situations, one mostly wants to confirm that the concentration drops off exponentially, so it may be sufficient to take some arbitrary number of the random locations and do profiles there. If the profiles are to be used in orcbe iitty statements or in significance tests, they should be randomly spread through the areas of interest. Where profiles are to figure in clean-up criteria, we probaily will need to try to design some specific plan. At present, we assume cnough will be taken to provide an initial notion of the situation for various islands and groups of islands. Obviously double-sampling should be useful in this case. Again, we might take more than seem needed and decide on wet chemistries on the basis of an initial set of analyses. Thinking ahead to the prospect of an actual clean-up operation, it seems to me that some experiments with compositing soil samples ought to be con- sidered. My arguments are as follows: I hope that it will be possible to set up the criteria on a probability basis. That this is necessary follows from the skewed nature of the frequency distributions. If these hold then there will always be some probability that any given criterion will be exceeded by a small (in some cases, vanishingly small) number of plots. One possible scheme is to subdivide the area into rectangles (or squares) of a size convenient for e.g., earth removal, and to take a sizable number of samples from each such bleck in turn, and to combine these into several composites and analyze the composites. If the mean of the composites exceeds a pre- determined value, then the plot is slated for clean-up. Such a procedure, perhaps using double-sampling, might reduce the number of samples that have to be analyzed. The immeuiate problem is that we know very little about the behavior of the compositing operation, and I'd hesitate to actually propose such a scheme without some data as to how well compositing works in practice. I hope that the above discussion will give you some background as to our notions of tne statistical aspects. We recommend that the entire sampling scheme be put together as soon as possible. Numbering the samples in order of the draw will permit such changes as become necessary in terms of field conditions. If the whole scheme is laid out in advance, there should then