Large changes in quantitative traits through small shifts in allele frequencies at many genetic loci driven by natural selection lie at the heart of the neo-Darwinian synthesis [1,2] but are hard to fully demonstrate in the wild [3-6]. The spread of a natural pathogen into a new host population containing standing genetic variation for quantitative resistance provides an opportunity to demonstrate such shifts occurring in real time [7-10]. The invasive fungal pathogen (Hymenoscyphus fraxineus) causes ash dieback disease in European ash (Fraxinus excelsior) populations. The disease is frequently lethal [11,12], but larger trees persist longer [11,13], often with reduced reproductive output [13]. Here, using whole genome resequencing of an infected natural ash tree population in England, we measure allele frequencies at 7985 single nucleotide polymorphisms (SNPs) previously associated with resistance to the pathogen [14]. The average genetic merit for resistance of trees established after the start of the epidemic, calculated as Genomic Estimated Breeding Values (GEBV) from the 7985 loci, was higher than that of related adults from the pre-epidemic generation. We estimate that, to produce a GEBV shift of this magnitude, would require truncation selection eliminating at least 13% of the juvenile population. Even this rapid rate of change under natural selection may be insufficient for evolutionary rescue [15] of European ash populations, but change could be accelerated by a breeding programme informed by genomic selection. To this end, we show that for predicting the health of their offspring, the GEBVs of parent trees perform better than a phenotypic assessment of the ash dieback damage parent trees have suffered.