We studied the spatial variability of soil temperature in two square grids, small and large, on a bare field, on silt loam soil located in Felin near Lublin. The soil temperature was measured every 1 m over a 3 by 3 m area for the small square grid and every 10 m over a 30 by 30 m area for the large square grid and at various soil depths. The spatial structure of soil temperature data was examined by semivariograms and classical statistics. Structural analysis of semivariograms for some depths consisted of pure nugget effect for different lag distances. This implies that the data are independent for our sampling scale and statistical properties could well be characterized by the mean and variance. Except for several soil depths which exhibited a pure nugget effect, the temperature observations were found to be spatially interdependent. The spatial structure was found at two scales: 18 m for depths of 2 -25 cm and 4.5 m for depths of 50-100 cm.