Previous studies have shown that approximately 40% of patients with myelodysplastic syndrome (MDS) and anaemia respond to treatment with human recombinant granulocyte-CSF (G-CSF) plus erythropoietin (epo). The present study was designed to investigate pre-treatment variables for their ability to predict erythroid responses to this treatment. 98 patients with MDS (30 RA, 31 RARS, 32 RAEB, five RAEB-t) were treated with a combination of G-CSF (0.3–3.0 μg/kg/d, s.c.) and epo (60–300 U/kg/d, s.c.) for at least 10 weeks. Minimum criteria for erythroid response was a 100% reduction of red blood cell (RBC) transfusion need or an increase in haemoglobin level of 1.5 g/dl. 35 patients (36%) showed responses to treatment. Medium duration of response was 11–24 months. In multivariate analysis, serum erythropoietin levels and initial RBC-transfusion need retained high statistical significance (P < 0.01). Using pre-treatment serum epo levels as a ternary variable (< 100, 100–500 or > 500 U/l) and RBC transfusion need as a binary variable (< 2 or 2 units per month), the analysis provided a predictive score for erythroid response. This score divided patients into three groups: one group with a high probability of erythroid responses (74%), one intermediate group (23%) and one group with poor responses to treatment (7%). This predictive scoring system could be used in decisions regarding use of these cytokines for treating the anaemia of MDS, both for defining patients who should not be given the treatment and for selecting patients for inclusion in prospective trials.