The last resort for conservation of rare tree populations in refugial areas under high risk of climate driven extinction may be ex situ conservation and assisted translocation. Although such actions require detailed knowledge about the spatial scale and heterogeneity of the within-population distribution of genetic diversity, it is still unknown whether fine-scale spatial genetic structure (FSGS) is present in refugial populations of forest trees. In order to address this issue, we carried out the first whole-population genetic characterisation of a small and isolated refugial population of the IUCN red-listed Serbian spruce [Picea omorika (Pan?.) Purk.] from the Balkans. All 418 adult individuals were georeferenced and genotyped at nuclear EST-SSRs and at a mitochondrial (mtDNA) locus. Spatial autocorrelation analyses provided only a simplified description of FSGS, which is concordant with findings in wind-pollinated species with limited seed dispersal. However, Bayesian analysis revealed three heterogeneous, highly differentiated (pairwise G’ST > 0.3), and spatially localised sub-populations showing only partial overlap with the distribution of mtDNA haplotypes. Such complex structure in only 0.34 ha, resulting mainly from historical events, restrictions to gene flow and high local density, was undetected in previous work based on more traditional sampling schemes for population genetics surveys. We demonstrate the usefulness of sampling schemes leaning towards a whole-population genetic characterisation in mining the finest characteristics of FSGS, and argue that our understanding of genetic structuring in highly heterogeneous refugial regions at both macro- and micro-scales is still rather limited and often oversimplified. This has severe implications on conservation of plant biodiversity from these regions in terms of responses to global climate change.