In this paper we compared the accuracy of different samples (minus the sample with the NN1 edge correction and buffer zone, plus sample and modified cluster sample) to determine indices of spatial structure in beech forests in Serbia, which can be used in regular forest inventory. The research was conducted in 11 forest stands of different structure at the territory of Central Serbia. Field data used for the assessment of spatial structure indices were collected across the examined stands on the sample plots of different sizes, using systematic grid, with the square layout ofsample plots (100x100 meters). The study was conducted on 242 sample plots. For accuracy comparison of different sample types, five indices were used: Uniform angle index, Mean directional index, Species mingling index, DBH differentiation and DBH dominance indices. One-way ANOVA showed that at the stand level there were no statistically significant differences between mean values of indices obtained by different sample types. However, Simple linear correlations confirmed for most indices that if the spatial structure is to be determined with accuracy at the level of sample plots (in this case, a circular fixed radius plot of 5 acres) it is necessary to use a plus sample during data collection. For species mingling and DBH differentiation indices, Simple linear correlations show that both minus sampling with edge correction buffer zone and modified cluster sampling with somewhat less accuracy can be used to a certain extent. Minus sampling with NN1 edge correction is not practical for use on sample plots of this size in beech forests in Serbia.