Population genetics analysis is crucial for understanding the degree of linkage disequilibrium in a germplasm collection for association mapping and breeding strategies. Here, we report population structure patterns and genetic differentiation of coffee germplasm collections in China using 19 ISSR markers. A total of 129 unique alleles were detected, with an average of 6.8 alleles per ISSR. The estimated average Nei's gene diversity (H_e) index for all tested accessions was 0.49. A clustering pattern correlating with predefined species origin, rather than isolation by distance, was observed for all the tested accessions. C. arabica is highly differentiated from C. liberica and C. canephora, with F_st values of 0.33 and 0.21, respectively. By contrast, C. liberica and C. excelsa Chevalier showed a close genetic association, with the F_st value of 0.04. Significant differentiation was also detected within C. arabica accessions, which were separated in three genetic subgroups supported by pairwise (F_st) index and Nei's standard genetic distance tests. The F_st outlier test for deviation from expected distribution between F_st and expected heterozygosity identified five putative ISSRs under selection, with two, UBC864 and UBC842, showing signals of positive selection. Our study provides useful information for identifying core collections for genetics studies, breeding programs, and development of future germplasm management strategies.