Abstract: Soil moisture content as an important variable in the land surface system, has significant influences on other components of hydrological cycles, and it is a key parameter in many hydrological models. Previous studies have shown that soil moisture in field conditions is highly variable in time and space. Therefore, knowledge of the characteristics of its variability is essential for understanding water flow processes in the field and for more accurate simulation of water movement in the field, and thus helpful to develop precision irrigation management practices. In this study, soil moisture contents were measured in a 7.6 hm2 vineyard located in the arid region of Northwest China during early May to later September in three consecutive years (2012-2014) using time domain reflectometry (TDR). Measurements were taken 17, 16 and 19 times for these years, respectively. Soil moisture data were collected for three soil layers (0 ~ 20, 20 ~ 40 and 40 ~ 60 cm) on a regular grid of 25 m × 25 m with additional points arranged near some grid nodes. There were 135 points for 2012 and 147 points for the other two years. The spatial structure of soil moisture were described using a variogram model, and evolutions of the fitted model parameters including the range and degree of spatial dependence (ratio of nugget to partial sill) were analyzed within the same year and between the three different years. In addition, possible relationships of the model parameters versus mean soil moisture were investigated for the three years, and simple linear regressions were conducted between the range and degree of spatial dependence in terms of mean soil moisture. The main results were: (1) Soil moisture generally had a wide range, and the range tended to be wider for the surface soil than the deeper soil in a wet year while the trend was opposite for a relatively dry year; (2) Higher degrees of spatial dependence were found for surface soils of a dry year and deeper soils of a wet year, and the range and degree of spatial dependence were increasing with increasing soil moisture, indicating that higher overall soil moisture in soil profile generally lead to greater correlation length but also often intensify the random component of overall spatial variability. The results of this study would help better describe and simulate the spatial distributions of soil water content in field conditions, and thus help develop site-specific precision irrigation management practices.