Soil samples different in scale were taken, and organic matter (O.M.), total nitrogen (Total N) and available phosphorus (Avail. P) were measured to study the feasibility and adaptability of GeoStatistic in analyzing soil spatial variability. Results indicated that soil O.M., avail. P and total N appeared in normal symmetrical distribution in samples large in scale but in samples medium or small in scale, they showed a slight skew distribution. The C.V. of different soil nutrients ranged between 0.17~0.27, showing a significant spatial variability, namely, when large in scale, soil spatial variability is low and when medium or small in scale, the variability is high (with small C.V. and kurtosis). A typical semivarigram structure was observed in samples of all scales, showing that the Geostatistic method works in analyzing soil nutrient spatial variability in samples of all scales. In samples small in scale soil nutrients have strong spatial autocorrelation (nugget value is among 0~0.17), and soil O.M. content constant spatial autocorrelation, while in samples medium or large in scale, the soil nutrient spatial autocorrelation is medium, but weak with soil Avail. P. Spatial interpolation and cross-validation showed that spatial prediction data and measured data fitted very well with the spherical model, and that no matter whether large, medium or small in scale, spatial analysis and prediction with the Geostatistical technology is a useful tool in soil spatial variability analysis and precision fertilization. The precision of prediction is higher in small scale than in large or medium scale. Among soil O. M., soil total N and soil available P, the prediction of soil O. M. has the highest precision.