strata combined (n = 120). This global approach (ignoring strata) in logarithmic scale seems appropriate here, but since 145 data are now available, and for this paper to be self-contained, we redo the analysis. The overall correlation between log(Pu) and log(FIDLER) from 145 pairs is p = 0.93. The regression equation is (see Figure 5) log(Pu) = 1.287 log(FIDLER) + 0.096 , (2) (Pu in uCi/m*, FIDLER in 103 cpm). It is interesting to comment on the fact that the slope 4 = 1.287 is greater than 1. Indeed, if the gamma Tay count recorded by the FIDLER is proportional to the Am concentration in a sample, and if the ratio R = Pu/Am is approximately constant, then FIDLER counts are proportional to Pu concentrations and the slope should be 1. The larger value found is explained by the difference in support between a soil sample analyzed for Pu--a 10~gram aliquot from the »v 700 grams of soil collected within the 5~inch-diameter sampling ring--and the much larger area integrated by a FIDLER reading. If the measurement errors effect is not overwhelming, it is to be expected that o2 << OF and since p is near 1, the regression slope 4 = Pd,/ Oo, is greater than 1. Further evidence of this interpretation is found using the data of Lines 1 and 2. The variances of log FIDLER counts are: LINE 1 (60 readings) Contact readings LINE 2 (25 readings) 0.100 Aboveground readings 0.342 0.087 0.062 As the area integrated by aboveground measurements is larger than the surface ones, these results are in the right direction, i.e., the variance of the aboveground readings is smaller than for the surface readings. The residual variance of the regression is 0.1313. We will not attempt to devise confidence intervals for the regression coefficients, for we . believe that the basic assumptions of independence on which classical regression theory is built are violated here. Looking carefully at the plot in Figure 5, one can distinguish two populations: one for which FIDLER > 5,000, and the other for which FIDLER < 5,000. Doing separate studies for these two subgroups, it is found that a 1. For FIDLER > 5,000, p = 0.89 and the regression is log(Pu) = 1.293 log (FIDLER) + 0.092 , (96 pairs) which is practically the same as that derived from the pooled samples. This circumstance is reassuring, as it is important to have a good regression where the activity is high. It is also 372