The estimation of the sample semivariogram and the selection of appropriate theoretical semivariogram model is the basic stage in the structural analysis of regionalized variables and spatial interpolation. However, due to the special characteristic of regionalized variable that there is only one realization at one sampling point, there exists uncertainty of the sample semivariogram calculated from only one realization of the stochastic field. As a result, assessment of sampling variance of the sample semivariogram becomes an important topic in the application of Geostatistics as it gives the magnitude of uncertainty associated with semivariogram estimates. Using soil water content data at the field scale, Jackknife variogram estimator and variance were calculated with the Jackknife method which is a parameter estimation technique developed by Quenouille to reduce bias involved in parameter estimate. Jackknife variance is used to qualify the magnitude of uncertainty associated with the sample semivariogram estimates. The uncertainty of the sample semivariogram for the soil water data were analyzed quantitatively and the confidence intervals of every lag distance with confidence level of 95% were given. Then confidence limits with a given confidence level for all lag estimates simultaneously were calculated using Bonferroni method for rectangular intervals. Finally, the inverse of Jackknife standard deviation at different lags were used as weights in weighted least square regression (WLS) to fit the spherical model, Gaussian model and exponential model, and the sensitivity of the WLS to initial value of parameters was analyzed. Exponential model was recommended as the theoretical semivariogram model according AIC.