A study on spatial variability of soil moisture for micro-scales was conducted in a kiwi field in Yangling region. The mesh generation was formed based on the separation distance of 8 m, 2 m and 0.5 m in the 40 m×40 m field. The spatial distribution and variation of soil moisture in 0~60 cm soil layers under different sampling densities were analyzed based on classical statistics and geo-statistics theories. Results showed that as for three micro-scales of 40 m×40 m (L), 8 m×8 m (M) and 2 m×2 m (S), the variability of soil moisture in 0~60 cm soil layers was weak and moderate in horizontal direction, and it decreased with the decrease of scale and the increase of depth. Meanwhile, the variability of soil moisture in 0~60 cm soil layers was also weak and moderate in vertical direction. A strong and significant spatial correlation was found in soil moisture in these three scales, and the semi-variance models for soil moisture varied with scales. This meant that semi-variance models for soil moisture in L scale could be simulated by spherical and exponential models; semi-variance models for soil moisture in M scale could be simulated by linear model; and semi-variance models for soil moisture in S scale could be simulated by Gaussian, exponential and linear models. Besides, the reasonable sampling amount in L scale was less than actual amount, but the reasonable sampling amounts in both M and S scales were more than actual amount. And as for L, M and S scales, the reasonable sampling amount in 0~30 cm layer was more than actual amount, but the reasonable sampling amount in 30~60 cm layer was less than actual amount. These results indicated that a reasonable layout of sampling point needs to be further optimized. In Yangling region, a significant difference of the spatial distribution of soil moisture could be found in the north and south of local kiwi field due to topography. These results could provide some guidance for irrigation and water control in precision agriculture.