Summary. We have published previously a prototype of a decision model for anaemic patients with myelodysplastic syndromes (MDS), in which transfusion need and serum erythropoietin (S-Epo) were used to define three groups with different probabilities of erythroid response to treatment with granulocyte colony-stimulating factor (G-CSF) + Epo. S-Epo ≤ 500 U/l and a transfusion need of < 2 units/month predicted a high probability of response to treatment, S-Epo > 500 U/l and ≥ 2 units/month for a poor response, whereas the presence of only one negative prognostic marker predicted an intermediate response. A total of 53 patients from a prospective study were included in our evaluation sample. Patients with good or intermediate probability of response were treated with G-CSF + Epo. The overall response rate was 42% with 28·3% achieving a complete and 13·2% a partial response to treatment. The response rates were 61% and 14% in the good and intermediate predictive groups respectively. The model retained a significant predictive value in the evaluation sample (P < 0·001). Median duration of response was 23 months. Scores for global health and quality of life (QOL) were significantly lower in MDS patients than in a reference population, and fatigue and dyspnoea was significantly more prominent. Global QOL improved in patients responding to treatment (P = 0·01). The validated decision model defined a subgroup of patients with a response rate of 61% (95% confidence interval 48–74%) to treatment with G-CSF + Epo. The majority of these patients have shown complete and durable responses.