We used the random subsampling approach based on the empirical data to identify the representative sample size for accurate estimates of allele frequencies within a population. The empirical data consisted of 12 nuclear microsatellite marker scores for 400 individuals sampled within 1 ha area in a representative natural stand of Scots pine. For each sample size, 100 resampled subsets were randomly drawn (without replacement). The sample size, at which 95 % of the resampled subsets contained all the alleles at a given frequency present in the empirical data set, was considered as a 95 % probability of sampling these alleles. The resampled subsets were also used to calculate main genetic diversity parameters and their variances to be used as a measure of accuracy of sampling. The results showed that at the 95 % probability level, the sample sizes of 20-25 and 65-80 individuals were large enough to capture all the alleles at frequency above 0.05 and 0.01-0.05, respectively. 300-350 individuals were required to sample the alleles at frequencies below 0.01 at the 95 % probability. The upper bound of the sample sizes was required for the loci exhibiting high He values (>0.80).