A number of soil sampling procedures (stratified random, zigzag, strip and regular grid) were compared with the traditional random soil sampling and composite soil sampling. Soil samples were collected from an area of about 3. 5 hectare at the Agricultural Research Institute. Dera Ismail Khan and analyzed for lime content, electrical conductivity, AB-DTPA extractable P and NH4OAc extractable K. The data were analyzed statistically using classical and geostatistical techniques. The mean values of different soil properties were the lowest in the soil samples collected by simple random method. For all the sampling methods, variability (CV) was minimum for lime content and maximum for AB-DTPA extractable P. Zigzag, strip and regular grid system at a spacing of 25 m x 25 m were consistently better, as indicated by the ratio, method CV/random CV, than random distribution method. The composite soil sample values of soil properties were not found to be the correct estimates. Regular systems required relatively consistent number of samples to accurately estimate the mean value of a soil property. Geostatistical analysis indicated the spatial distribution of lime content and random distribution of other soil properties. The spherical model for lime content semivariogram had a range value of 56 m.