Generally, effect of fragmentation per se on biodiversity has not been separated from the effect of habitat loss. In this paper, using nDNA and cpDNA SSRs, we studied genetic diversity of Castanopsis sclerophylla (Lindl. & Paxton) Schotty populations and decoupled the effects of habitat loss and fragmentation per se. We selected seven nuclear and six cpDNA microsatellite loci and genotyped 460 individuals from mainland and island populations, which were located in the impoundment created in 1959. Number of alleles per locus of populations in larger habitats was significantly higher than that in smaller habitats. There was a significant relationship between the number of alleles per locus and habitat size. Based on this relationship, the predicted genetic diversity of an imaginary population of size equaling the total area of the islands was lower than that of the global population on the islands. Re-sampling demonstrated that low genetic diversity of populations in small habitats was caused by unevenness in sample size. Fisher's α index was similar among habitat types. These results indicate that the decreased nuclear and chloroplast genetic diversity of populations in smaller habitats was mainly caused by habitat loss. For nuclear and chloroplast microsatellite loci, values of F_ST were 0.066 and 0.893, respectively, and the calculated pollen/seed dispersal ratio was 162.2. When separated into pre-and post-fragmentation cohorts, pollen/seed ratios were 121.2 and 189.5, respectively. Our results suggest that habitat loss explains the early decrease in genetic diversity, while fragmentation per se may play a major role in inbreeding and differentiation among fragmented populations and later loss of genetic diversity.