Amentotaxus is an ancient lineage of gymnosperm genus. Based on the Tertiary fossil evidence, the ancient lineage was once widespread in the Northern Hemisphere but range contraction in Pleistocene and Holocene. Currently, the distributional range is restricted to southwestern China, Vietnam and Taiwan. In this study, the 15 polymorphic microsatellite primers were designed from A. formosana H. L. Li, and were used these primers for genetic research of species level and population level. In species level, we tested the transferability and polymorphism in four species, A. argotaenia (Hance) Pilger, A. formosana H. L. Li, A. poilanei (Ferr&#x00E9; & Rouane) D.K. Ferguson and A. yunnanensis H. L. Li were evaluated. In population level, we evaluate the genetic variation and genetic structure on two habitats including Mt. Chachayalaisham and Jinshueiying for the delimited of relevant genetic units and to establish useful conservation strategies. In species level, the 15 microsatellite primers had 100% successfully amplified fragments in these three species, also showing high polymorphisms (PIC=0.25-0.61). In population level, the genetic diversity analysis in the two population of A. formosana show that expected and observed heterozygosities were ranged from 0.52 to 0.60 and 0.28 to 0.36, respectively. In addition, 15 loci were deviated from Hardy-Weinberg equilibrium (P<0.001) and the inbreeding coefficient were displayed positive (FIS=0.48368; P <0.05), revealing the genetic fixation causing by inbreeding. The analysis of molecular variance (AMOVA) revealed that high genetic variation with population (83.85% and 52.39%), and also showed high levels of population differentiation (FST =0.10972-0.16155; P <0.05). IMa revealed low migration rate (1.13¡Ñ10-6-1.15¡Ñ10-4), also showed small effective population size (Cha: 77.36-1830.67; Jin: 21.41-506.67) and large ancestral population size (10496.8-248424.17), inferring the A. formosana undergo significant population declined. Using Bayesian clustering algorithms and Markov chain Monte Carlo (MCMC) iterations to distinct genetic units and make assignments in the programs SAMOVA¡BSTRUCTURE and INSTRUCT, the best clustering was occurred at K=2, and found 11 and 9 distinct genetic groups in Mt. Chachayalaisham and Jinshueiying based on GENELAND analysis, respectively. We correlated the distinct genetic units and the age structures to evaluate the individuals with unique genotype for seed source to maintain the maximum genetic diversity of A. formosana.