Spatial distributions of a pollutant in contaminated soil are essential for risk assessment and soil remediation. Recently, the kriging technique is frequently used in spatial interpolation of pollutant concentrations to delineate contaminated area. However, high misclassification rates of hazardous area based on kriging estimated values of pollutant concentration will happen at the locations, where pollutant concentrations are close to the regulation thresholds for declaring hazardous area and where the first sampling density is low. Thus, in order to reduce the misclassification rate of delineations, an adaptive sampling approach, called the Adaptive Cluster Sampling (ACS) based on both the first sampling density and the regulation threshold of pollutant concentrations, was proposed to improve kriging estimation and thus to reduce the misclassification rates in this study. For assessing the feasibility of the adaptive sampling approach, a comparison of the adaptive cluster sampling and Simple Random Sampling (SRS) was carried out in simulation. A data set of soil Ni concentrations in a heavy-metal contaminated site in the Changhua County, Taiwan, was used for illustration. The simulation results showed that the additional samples with soil Ni concentrations close to the regulation threshold (Ni = 200 mg kg-1) were drawn in the ACS but not in the SRS. Simultaneously, the sampling configuration of additional samples in the ACS was on the locations where the first sampling density is low, but it was spreading randomly on the site in the SRS. The precision of kriging estimation of soil Ni concentrations based on the observations sampled in the ACS was thus relatively higher than that based on the observations sampled in the SRS. Compared with using the SRS, using the ACS proposed could reduce the misclassification rate of hazardous area delineation based on kriged soil Ni concentrations.