The sampling sites of the French Soil Monitoring Network (FSMN) are selected by systematic random sampling (SY). It consists of a 16 x 16 km grid, leading to a total of about 2200 sites.. SY leads to good spatial coverage, i.e. the sites are uniformly spread over France, thereby enhancing the precision of design-based estimates of spatial means and totals of various trace elements among other soil properties. Besides, SY is a suitable sampling design for spatial mapping using e.g. kriging. Therefore, SY is a flexible sampling design: its samples can be used both for design-based estimation of means and totals, and for mapping. Design-based estimation of a spatial mean or total from SY samples is straightforward. However, this is not the case for the sampling variance of the estimated mean or total. An unbiased estimator of this sampling variance does not exist. Three different approaches may be considered to approximate the sampling variance. First, a simple approximation is to calculate the sampling variance as if the sample were a simple random sample. In general this procedure over-estimates the sampling variance. A second approximation is to treat the SY sample as a stratified simple random sample. In this approach the SY locations are clustered into pairs of locations on the basis of their spatial coordinates. In general with this approximation the over-estimation of the sampling variance will be less serious compared to the first approximation. A third option is to predict the sampling variance from a variogram. In this approach the SY sample is used to calibrate the variogram. The sampling variance can then be predicted from mean semivariances within grids and within the study area. A last approximation is to calculate the sampling variance as if the sample were a simple random sample and to multiply this first approximation by a correction factor derived from Moran’s spatial autocorrelation statistic I. In this work, we first explored these variance approximations in a simulation study. A map of NDVI as obtained by MODIS was used as reality. The map was sampled a large number of times by SY, using the 16 km grid-spacing of FSMN. For each SY sample the means of NDVI within broad parental material units of the soil map of France were estimated. The experimental sampling variances thus obtained were compared with the approximated sampling variances Secondly, the data of the first campaign of the French Soil Monitoring Network were used for design-based estimation of the means of trace elements (Cd, Co, Cr, Cu, Pb, Zn) for the above-mentioned map units and to approximate their sampling variances.