Detecting signatures of selection in tree populations threatened by climate change is currently a major research priority. Here, we investigated the signature of local adaptation over a short spatial scale using 96 European beech (Fagus sylvatica L.) individuals originating from two pairs of populations on the northern and southern slopes of Mont Ventoux (south-eastern France). We performed both single and multi-locus analysis of selection based on 53 climate-related candidate genes containing 546 SNPs. FST outlier methods at the SNP level revealed a weak signal of selection, with three marginally significant outliers in the northern populations. At the gene-level, considering haplotypes as alleles, two additional marginally significant outliers were detected, one on each slope. To account for the uncertainty of haplotype inference, we averaged the Bayes Factors over many possible phase reconstructions. Epistatic selection offers a realistic multi-locus model of selection in natural populations. Here, we used a test suggested by Ohta based on the decomposition of the variance of linkage disequilibrium. Over all populations, 0.23% of the SNP pairs (haplotypes) showed evidence of epistatic selection, with nearly 80% of them being within genes. One of the between gene epistatic selection signals arose between an FST outlier and a non-synonymous mutation in a drought response gene. Additionally, we identified haplotypes containing selectively advantageous allele combinations which were unique to high or low-elevations and northern or southern populations. Several haplotypes contained non-synonymous mutations situated in genes with known functional importance for adaptation to climatic factors