STUDY QUESTION: Do different dosages of metformin account for different clinical and biochemical outcomes in women with polycystic ovary syndrome (PCOS) and do basal anthropometric and metabolic characteristics of the patients provide any indications regarding the dose required to reach the target effect? SUMMARY ANSWER: Different doses of metformin exerted the same effects on clinical, biochemical and metabolic parameters in patients affected by PCOS. WHAT IS KNOWN AND WHAT THIS PAPER ADDS: Since the insulin-sensitizing agents came into use in the management of PCOS, metformin has shown a positive benefits-risks ratio. Nonetheless, therapeutic schedules are not well standardized. This is the first study which systematically analyses the effect of different doses of metformin on clinical, hormonal and metabolic features of PCOS. On the basis of our results, higher doses are no more effective than lower doses. DESIGN: A multicentric cohort prospective study. A total of 250 PCOS women were enrolled, 49 lost to follow-up. Menstrual cyclicity, hormonal assays, oral glucose tolerance test, lipid profile and ultrasonographic pelvic examination were evaluated at the baseline and after 6 months of metformin treatment at different doses (1000, 1500 and 1700 mg). PARTICIPANTS AND SETTING: A total of 201 PCOS patients completed the study without protocol violations in three university hospitals: seventy-three patients from Centre A (treated with metformin 500 mg twice a day), 60 patients from Centre B (treated with metformin 500 mg three times a day) and 68 patients from Centre C (treated with metformin 850 mg twice a day). MAIN RESULTS AND THE ROLE OF CHANCE: Metformin exerted an overall positive effect on the clinical and endocrine-metabolic features of PCOS. The degree of these effects was independent of the administered dosage in every range of basal body mass index (BMI). When patients were stratified according to their insulinaemic status, scattered inter-doses differences were found in some of the outcome measures. Patients who exhibited an increase of >2 menstrual cycles/year were considered as responders to treatment. Responders had a higher basal BMI than non-responders and showed a greater reduction in plasma testosterone levels after metformin treatment, but other outcome measures did not differ significantly. Total insulin secretion in the 180 min following the glucose tolerance test before metformin treatment (basal AUC-I) was significantly correlated with the decrease in insulin secretion induced by metformin in both the whole group and in responders, but only correlated with the variation in the number of cycles in responders. BIAS, CONFOUNDING AND OTHER REASONS FOR CAUTION: The different doses were administered in different centres, and between-centre variation is a potential confounding factor. GENERALIZABILITY TO OTHER POPULATIONS: The paradigm of using the minimum effective dose of metformin could be pursued in other pathological conditions characterized by insulin resistance.