Geostatistical concepts were used to evaluate the spatial variability of soil physical properties. A 400-acre cotton field was sampled on a 300 feet grid basis, taking 209 soil profiles at 3 depths. The spatial analysis of the soil texture, bulk density, water retension, and hydraulic conductivity data indicate high variability across the field. The semivariogram models show minimum nugget, sill and presence of spatial structure beyond its original sampling distance, except the saturated hydraulic conductivity with high nugget (109 cm day^-1), and high sill (390 cm day^-1) values. This large residual variation over small distances, could be due to wide sample spacing. To reduce the nugget effect a shorter grid sampling scale may be adopted for measurement of saturated hydraulic conductivity. The GOSSYM/COMAX-Cotton model was used on the same spatially variable sites to determine if the crop can be managed based on dominant soil types or could be managed site specifically. GOSSYM/COMAX was first run on selected 26 sites, consisting of 6 major soil series which represent 97% of the total soil series of the field. Then common soil series, physico-chemical properties were averaged to use as input in the model. The individual and averaged predictions were compared with each other and with the actual crop growth parameters and yield. The individual and averaged predictions for plant height, numbers of nodes, and yield were in good agreement. A significant correlation (r = 0.87) was found between individually predicted yields for soil series and actual yield.