Erythropoietin (epo) can be used to improve the anaemia of patients with myelodysplastic syndromes (MDS), but the efficacy is relatively low and the treatment is expensive. So far, no individual clinical trial has been sufficiently extensive to provide a basis for a decision model for the use of epo in MDS. This meta-analysis included 17 original articles with a total of 205 patients with MDS who had been treated with epo. 33 patients (16%) showed a significant response to treatment. Patients with refractory anaemia with ring sideroblasts (RAS) showed a lower response rate than all other patients (7.5% v 21%, P = 0.010). The difference in response rate between patients with and without transfusion need was also highly significant (10% v 44%, P < 0.001). The serum level of epo was significantly lower in the responding patients, but this parameter on its own could not be used to identify patients with a favourable response. FAB group (RAS versus others), transfusion need and s-epo (>/< 200 U/l) were combined in a model to provide information about the probability of response in different groups of patients. Patients with no transfusion requirement and MDS other than RAS showed a response rate of > or = 50%, irrespective of their serum level of epo. In patients with RAS and s-epo > 200 U/l, no response was observed. In the remaining groups the response rates varied between 9% and 33%. This meta-analysis shows that the efficacy of epo in MDS in general was low, but that groups of patients with an acceptable response rate could be identified.