Dispersal studies provide valuable information in agricultural field trials, e.g. cultivar experimentation of genetic modified crops. The amount of dispersal in plant populations can be indirectly estimated from the observed spatial patterns or autocorrelations in the data. Positive spatial autocorrelation [i.e. nearby observations tend to be more similar than distances ones] is assumed to result from any kind of spatial process, such as pollen flow or seed dispersal. The studies of spatial population genetic structure provide better understanding of local breeding and trial managment. Since in a short distance scale the spatial structure could be weak, efficient statistical approaches are needed to detect genetic autocorrelations in field experiments. Multivariate analysis based on microsatellite data, which increased significatively the numbers of alleles and loci containing genetic information, improved the power to detect spatial trends. The spatial process is often described in terms of the maximum distance to which such structure extends. Variogram modelling, from field-genetic distance plots, summarizes the spatial genetic structure within a plant population under isolation-by-distance. The variogram can be interpreted as a distance-dependent estimate of the population variance under spatial structure. However, parametric geostatistic analysis requires prior knowledge and assumptions that could be difficult to meet in field trials. In this work we use a non-parametric variogram approach to analyze multilocus-multiallele microsatellite data studied in space. The empirical variogram was fitted by means of local regression smoothing (LOESS) allowing estimation of genetic variability, and to assessment of the field distance beyond which observations are spatially uncorrelated. The non-parametric variogram was simpler to fit and represented an efficient statistical tool to analyze geo-referentiated field observations.