Background: Many diabetic patients are admitted to hospital, where care is costly and where there may be scope to improve efficiency. Aims: We analyse the costs and characteristics of diabetic patients admitted to English hospitals and aim to assess what proportions of cost variation are explained by patient and hospital characteristics. Methods: We apply a multilevel approach recognising that patients are clustered in hospitals. We first analyse the relationship between patient costs and their characteristics, such as HRG, age, gender, diagnostic markers and socio-economic status. We derive the hospital fixed effect and adjust for hospital characteristics such as number of patients treated, factor prices and number of specialties involved in diabetes care. We rank hospitals by their adjusted fixed effect, which measures the extent to which their costs vary from the average after controlling for patient and hospital characteristics. We conduct sensitivity analysis to alternative specifications including different sets of covariates and subsamples. Data: We use Hospital Episode Statistics and reference costs for all patients admitted to diabetes care for all English hospitals for the financial year 2005/06. Our sample includes 31.371 admitted patients in 148 hospitals. Results: Much of the variation in costs is driven by patient characteristics. Even so, around 8-9% of the variation in costs is across hospitals. Geographical variation in factor prices is significant for cost of diabetes care. The volume of patients, the number and diversity of specialties involved in caring for diabetics was rejected as significant cost drivers. . Conclusion: Health Resource Groups (HRGS) and diagnostic markers are significant patient-related cost drivers in diabetes care. Costs are not lower in hospitals with high volume of patients and where diabetes care is concentrated in few specialties.