Treatment with recombinant erythropoietin (EPO) can alleviate anaemia in patients with myelodysplastic syndromes (MDS). The present study, based on a long-term follow-up of 68 MDS patients (26RA, 16 RAS, 26 RAEB) treated with EPO alone, pinpoints pre-treatment variables associated with response induction, response duration and overall survival. Response, defined as an increase in haemoglobin >15gL1 or eliminated erythrocyte transfusion requirements, was observed in 22 of 66 (33%) evaluable patients. The median response duration was 15 (range 3-64+) months. Using univariate logistic regression models, responders displayed significantly lower baseline serum EPO levels (S-EPO), more often normal bone marrow blast cell content (RA/RAS vs. RAEB), normal cytogenetics and no need for erythrocyte transfusion. In a multiple logistic regression model, S-EPO (P=0.009), marrow blast content (P=0.031) and erythrocyte transfusion need (P=0.024) remained associated with response induction. The probability of response for a patient with S-EPO >50UL1, RA/RAS and no transfusion need was 0.79 (0.53-0.93, 95% CI). The median overall survival time from start of EPO treatment was 26 months, significantly longer for responders than for non-responders (49 vs. 18 months, P=0.018). Survival was also predicted by baseline S-EPO; patients with S-EPO >50UL1 (n=50) had a median survival of 17 months, as compared to 65 months for those with S-EPO >50UL1 (n=14, P=0.024). The international prognostic scoring system (IPSS) for MDS predicted survival (P=0.003) and progression to acute leukemia (P<0.001) but not response to EPO treatment. Furthermore, in a logistic regression model with S-EPO and IPSS, S-EPO (but not IPSS) was again a significant predictor for response (P=0.007). Our data facilitate the optimal selection of MDS patients suitable for EPO treatment and pinpoint S-EPO as a powerful predictor of response and overall survival in MDS.