The objectives of the present analyses were to assess the association between HbA1c levels and severe hypoglycaemia (SH, treatment with glucose i.v. or glucagon injection) and to identify predictors of SH in a prospective multicentre trial. The study population consisted of 636 insulin-dependent diabetic patients who had participated in a structured 5-day in-patient group treatment and teaching programme for intensification of insulin therapy (ITTP) in one of 10 hospitals and who were re-examined after 1, 2, 3, and 6 years including assessment of demographic, disease and treatment related parameters, diabetes-related knowledge, behaviour, and emotional coping. At baseline, age (mean +/- SD) was 27 +/- 7 years, diabetes duration 9 +/- 7 years and HbA1c 8.3 +/- 1.9 %. During the 6-year follow-up, the mean HbA1c value improved to 7.6%, and in patients with a diabetes duration of more than 1 year at entry into the study (n = 538) the incidence of SH decreased from 0.28 cases/patient/year during the year preceding the ITTP to 0.17 cases/patient/year. The patient group was divided into decile groups according to mean follow-up HbA1c values. In each group more than 230 patient years could be analysed. Groups with mean HbA1c values of 5.7, 7.0, 7.4, 7.7 and 8.9% had comparable risks of SH (0.15-0.19 cases/patient/year). In a logistic regression analysis, mean HbA1c during follow-up, a history of SH during the year preceding the ITTP, C-peptide level, emotional coping, carrying emergency carbohydrates (as assessed at the 1-year follow-up), and age at onset of diabetes were significant independent predictors of SH. The incidence of SH between centres varied between 0.05 and 0.27 cases/patient/year. In conclusion, in the present analyses no linear or exponential relationship between HbA1c and severe hypoglycaemia could be identified by using simple group comparisons. Applying complex regression analyses, various patient-related predictors of severe hypoglycaemia were identified.