This paper presents a new approach to predict soil properties and quantify uncertainty in the derived soil property maps over large areas using sparse and ad-hoc samples. According to the soil-landscape model, each soil sample contains corresponding relationships between soil and environment conditions. Under the assumption that the more similar the environmental conditions between two locations the more similar the soil property values, each sample can be considered as a representative (individual representativeness) over areas of similar environmental conditions. The level of representativeness of an individual sample to an unsampled location can be approximated by the similarity in environmental conditions between the two locations. Based on this “individual representativeness” concept and with the use of the Case-based Reasoning (CBR) idea, which solves new problems by referring to similar cases, soil property values at unsampled locations can be predicted based on their environmental similarity to the individual samples. Furthermore, the uncertainty associated with each prediction is related to the similarity and can thus be quantified. A case study located in Illy Region, Xinjiang, Northwest China, has demonstrated that the predicted map of soil organic matter of top layer is of good quality and the quantified uncertainty is positively correlated with prediction residual. This suggests that the approach can be an effective alternative for predicting soil property and reporting uncertainty in the resulting soil map over large areas with sparse and ad-hoc samples.