Ions occurring in the soil solution as well as exchangeable ions related to exchangeable complex are a reservoir of magnesium for plants. The content of magnesium forms important for plants can reveal a very high spatial variability of this element on a field scale. Determination of such spatial variability of elements is extremely important for the so-called precision agriculture, in which application of a fertilizer dose is by principle related to the actual deficit of elements, even in a small area. In order to evaluate parameters connected with the spatial variability of magnesium forms, the H2O (water) soluble (Mg-H2O), exchangeable (Mg-E) and available magnesium (Mg-A) forms were determined. The soil samples were taken in spring 2007 from the humic Luvisols horizon under winter wheat. Fifty soil samples were taken from the sites located in a square sampling grid (10 m x 10 m). The results were evaluated with the use of classical statistical methods as well as geostatistical calculations. The raster maps illustrating the spatial variance of determined nitrogen forms were drawn on the ground of semivariograms. The concentration of magnesium forms in the surface horizon of the soil showed significant differentiation: Mg-H2O ranged from 0.76 to 2.89 mmol(+) kg–1; Mg-E 1.69-8.06 mmol(+) kg–1 and Mg-A 28.50-91.40 mg kg–1. The data were analyzed statistically. Coefficients of variations equaled 31.6% for Mg-H2O; 30.5% for Mg-E and 24.9% for Mg-A. Analysis of dispersion showed the highest similarity to the mean value, reaching 1.24 mmol kg–1 Mg-H2O, which was confirmed by high kurtosis (8.73). The most flattened distribution was noted for Mg-E (– 0.39 kurtosis). Geostatistical calculations demonstrated that the analyzed magnesium forms did not occur in total dispersion in the soil mass, which was confirmed by high nugget variance values equal 0.423 (mmol kg–1)2 for Mg-E, 0.031 (mmol kg–1)2 for Mg-H2O and 83.42 (mg kg–1) for Mg-E. The highest participation of the nugget variance in sill variance (47.3%) was observed on the Mg-E semivariogram. The range of influence, defined as a maximum distance of correlations between point values, ranging 80.0-98.0 m, was higher than the real distance of soil sampling.