OBJECTIVES: To assess whether in people with poorly controlled type 2 diabetes (HbA1c>7.5%) improvement in HbA1c varies by ethnic and social group. DESIGN: Prospective 2-year cohort of type 2 diabetes treated in general practice. SETTING AND PARTICIPANTS: All patients with type 2 diabetes in 100 of the 101 general practices in two London boroughs. The sample consisted of an ethnically diverse group with uncontrolled type 2 diabetes aged 37-71 years in 2007 and with HbA1c recording in 2008-2009. OUTCOME MEASURE: Change from baseline HbA1c in 2007 and achievement of HbA1c control in 2008 and 2009 were estimated for each ethnic, social and treatment group using multilevel modelling. RESULTS: The sample consisted of 6104 people; 18% were white, 63% south Asian, 16% black African/Caribbean and 3% other ethnic groups. HbA1c was lower after 1 and 2 years in all ethnic groups but south Asian people received significantly less benefit from each diabetes treatment. After adjustment, south Asian people were found to have 0.14% less reduction in HbA1c compared to white people (95% CI 0.04% to 0.24%) and white people were 1.6 (95% CI 1.2 to 2.0) times more likely to achieve HbA1c controlled to 7.5% or less relative to south Asian people. HbA1c reduction and control in black African/Caribbean and white people did not differ significantly. There was no evidence that social deprivation influenced HbA1c reduction or control in this cohort. CONCLUSIONS: In all treatment groups, south Asian people with poorly controlled diabetes are less likely to achieve controlled HbA1c, with less reduction in mean HbA1c than white or black African/Caribbean people.