The objective was to evaluate the effect of the treatment of anemia with recombinant human erythropoietin (EPO) on thrombosis of the vascular access used for hemodialysis. The research design was a prospective cohort study comparing EPO-treated hemodialysis patients with a comparison group matched for type of vascular access, clinical center, and age. All patients commencing hemodialysis in the study centers between March 1988 and July 1991 were eligible if either a graft or fistula had been used as a first permanent vascular access. There were 64 matched fistula pairs and 38 matched graft pairs. There were more patients with a history of cardiovascular disease in the EPO group than in the comparison group for both fistulae and grafts, 34 versus 14% for the former and 37 versus 5% for the latter. There was no difference between EPO and comparison groups with respect to time to first thrombosis of fistula, 11.3 versus 10.6%, respectively, by thrombosis of grafts among those treated with EPO--33.6 versus 11.2% (P = 0.02). EPO treatment does not increase the probability of fistula thrombosis, but there is an association with an increased probability of graft thrombosis.