In conventional agriculture common practices may often lead to serious problems for soil which becomes too compacted so that water is hampered to penetrate the soil. Conservative agriculture should improve crop yields and resilience against drought and other hazard, while in the same time protecting and stimulating the biological functioning of the soil. Spatial heterogeneity is an important source of external variation which can influence the results of comparison between conventional and conservative agriculture. Geostatistics is an alternative approach to traditional statistics because it allows to account for spatial correlation. The aim of this study was to present a methodology to evaluate the effect of conservative agriculture on soil water content taking into account spatial autocorrelation. The research was carried out on the experimental farm of the CRA-Agronomic Research Institute, located in Foggia (41° N, 15° 40' E, 90 m above sea level), southern Italy. The soil is a deep, silty-clay Vertisol of alluvial origin, classified as fine, mesic, Typic, Chromoxerert (Soil Survey Staff, 1999). The climate is characterized by hot and dry summers and rains concentrated mostly in the winter months. A field trial on rainfed durum wheat was carried out on a 0.5-ha area according to a completely randomized block design with two treatments (conventional and conservative agriculture) and with three replications (block). Volumetric soil water measurements were made at 135 points with 15-cm long rods of a two-probe Trase System TDR (Time Domain Reflectometry) at the nodes of a regular grid of size 5 m x 5 m. The measurements were repeated at the same locations on seven times from 2003 to 2005. In order to assess the spatial relationship between soil water content and treatment we used an indicator approach (Journel, 1983) based on a prior coding of each observation as an indicator datum (0 or 1) according to a cut-off equal to the field average of soil water content at each measurement time. An additional indicator variable was the treatment type (categorial attribute). First, a Linear Model of Coregionalization was fitted to the matrix of direct and cross-variograms of the indicator variables. An ordinary indicator-cokriging was then applied to estimate the probability of exceeding the field average of soil water content at each measurement time. The spatial relationship (cross-variogram) between the two indicator variables was negative during the wet periods but positive during drought, which means that the probability to have water content greater than field average was higher in the plots submitted to conservative agriculture than in the traditionally managed ones during the dry periods. This results was also confirmed by the probability maps which showed the dual behavior of soil water content as a function of rainfall pattern. The proposed geostatistical approach has then proved the positive effect of conservative agricultural practices on soil water conservation during drought.