larger after iterating. This behavior results, most likely, from the difficulty of fitting a surface to the very large concentration observed near GZ without distorting grid estimates in nearby areas. Residuals in antilog form at sample collection points are usually smaller on the average than those from fits in the wuntransformed scale. Also, distortions in the antilog concentration surface due to the extreme data point near GZ appear to be minor. However, the correlation between observed data and residuals tends to be somewhat larger for the log and antilog fits than for untransformed data fits. Moreover, fitting in log scale does not, of course, eliminate fitting problems due to a lack of data in certain regions of the study site. Recall, for example, the apparently spurious contours south of GZ for both the untransformed and antilog fits due to insufficient number and placement of samples in that area (Figures 6-9). The plotting of observed data and residuals on estimated contour maps has proved useful in subjectively evaluating the estimated surface at other than data points. Efforts should be made, however, to try kriging techniques to estimate these concentration surfaces,* since this approach yields an estimation variance for each grid node estimate. Furthermore, if the underlying assumptions of kriging are satisfied, this method yields best linear unbiased estimates of concentration at grid nodes. ACKNOWLEDGMENTS I wish to thank Ms. Barbara Vinson of Battelle-Northwest for her excellent assistance in writing computer programs and performing so well a variety of other computing tasks associated with this paper. Special recognition is also due Ms. Mary Lou Lomon, who did such an excellent job in typing the manuscript. *See Delfiner and Gilbert (1978), this volume. 354