Genetic variation of forest coffee trees (Coffea arabica L.) from four regions of Ethiopia was investigated using inter-simple sequence repeat (ISSR) markers. A total of 160 individuals representing 16 populations were sampled. Eleven ISSR primers amplified a total of 123 fragments of which 31 fragments (25%) were polymorphic. Estimate of total gene diversity (HT), and the coefficient of genetic differentiation (GsT) were 0.37 and 0.81, respectively. This indicates that most of the variability is between populations than within populations. The partitioning of genetic variation into within and between populations based on Shannon's information index also revealed more differentiation between populations (0.80) than within populations (0.20). In the phenogram most of the coffee tree samples were clustered on the basis of their regions of origin but failed to cluster according to their respective populations, which could be attributed to the presence of substantial gene flow between adjacent populations in each region assisted by man in the process of transplantation or by wild animals such as monkeys, which eat the berries and defecate the seeds elsewhere. On the other hand, the inter-regional clustering of some coffee tree samples from Bale and Jimma regions could be due to the transport of coffee seeds across regions and their subsequent planting. Although ISSR markers detected lower polymorphic loci than previously reported results with random amplified polymorphic DNA (RAPD) markers on the same materials, it can be used as an alternative method for molecular characterization of C. arabica populations. The results may provide information to select sites for in situ conservation.