List members: The original question appears at the bottom of the message. Thanks go to Edzer Pebesma for his assistance. As far as the ordinary kriging routine, my improper specification of the range parameter was the culprit (see page 38 of the gstat User's Manual for the gstat convention). Edzer reminded me of the different conventions for specifying the range parameter with certain variogram models and referred me to the following link for the GSLIB convention: http://www.ai-geostats.org/FAQ_software_conventions.htm In a related issue that was not mentioned in my post to the list: The large sill value in my cmd file did not create a problem for the ordinary kriging routine however, it did cause the SGS routine to generate negative predictions in many areas of the site that I am working with. Some of the predictions were quite large and occurred in areas of the site where only large concentrations (>500 ppm) were measured. I tried different methods to avoid this problem including: reducing magnitude of data, standardizing the data, increasing the value of zero (see pages 41 and 87), increasing the amount of data used in the prediction and performing block simulation at different block sizes. I was able to prevent the prediction of negative values was to use the original data (i.e., original magnitude) with a reduced sill. However, I think this is unsatisfactory because, unlike kriging, the value of the sill does effect the magnitude of the predicted concentrations. Edzer suggested that I tried the normal score transform. This produced much better results. Thanks again, Bill Original Message: I am new to gstat. As a introduction to the software I have been trying to reproduce some results that I have obtained using GSLIB. Unfortunately, I have not been able to obtain the same results. One of the problems that I have had is gstat produces negative (in some cases very large) estimates for values that should all be positive (I am working with concentration data). I know that in some cases, negative values can be generated by the 'screening effect' by which large sample data receive a negative weight. However, this did not occur with GSLIB. I did suspect that the problem was caused by my specification of the variogram model. However, the model I am using is very simple. I have included the gstat command file and the GSLIB parameter file below. I have also tried the following with no appreciable difference in the results: 1) increased search distance to 400 feet 2) Using force option to ensure the minimum number of sample observations (4) were included in the interpolation 3) Tried changing minimum number of points to 2 4) Changed gstat block discretization from default (Gauss quadrature) to 4 x 4 regular block discretization Any ideas what I am doing wrong? I would also like to hear from others that have compared results from the two programs. Thanks, Bill