Abstract: Peach accessions conserved in National Fruit Germplasm Repository of Nanjing were used for establishing primary core collection. Three sampling strategies，classification，stepwise cluster and random，were compared using 630 accessions and 60 descriptors. The sampling scale was also tested with considering the sampling strategies. Stepwise cluster was found to be more suitable for peach than sampled by classification or random. By multiple comparisons，15% sampling ratio was shown to be the best after stepwise cluster. Differences were found among sixteen cluster models and among sampling methods after cluster. Clustered by Euclidean-Ward model and selecting accession with higher value of average degree of deviation was chosen as the best method. A primary core collection of 99 accessions with sampling ratio of 15.7% was established based on the optimal method and by adding absent phenotypes to obtain 100% ratio of phenotype retained（RPR）. The core collection was detected to be well represent the primal collection using Shannon-Wavear index（SI），variance of phenotype value（VPV），coefficient of variance（CV），as well as，detected by coincidence of mean and range for quantity traits.

Key words: peach, phenotype, primary core collection