Aims We assessed the association between different blood lipid measures and risk of fatal/nonfatal coronary heart disease (CHD), which has been less analysed previously in type 2 diabetes. Design, methods Observational study of 46,786 patients with type 2 diabetes, aged 30-70 years, from the Swedish National Diabetes Register, followed for a mean of 5.8 years until 2009. Baseline and updated mean low-density lipoprotein (LDL)-, high-density lipoprotein (HDL)-, non-HDL-cholesterol, and non-HDL-to-HDL-cholesterol ratio were measured. Results Hazard ratios (HR) for CHD with quartiles 2-4 of baseline lipid measures, with lowest quartile 1 as reference: 1.03-1.29-1.63 for LDL; 1.23-1.41-1.95 for non-HDL; 1.29-1.39-1.57 for HDL; and 1.31-1.67-2.01 for non-HDL:HDL, all p<0.001 except for quartile 2 of LDL, when adjusted for clinical characteristics and nonlipid risk factors. A similar picture was seen with updated mean values. Splines with absolute 6-year CHD rates in a Cox model showed decreasing rates only down to around 3mmol/l for LDL, with linearly decreasing rates to the lowest level of non-HDL:HDL. Non-HDL and HDL were independent additive risk factors for CHD risk. HRs per 1SD continuous decrease in baseline or updated mean HDL were 1.14-1.17 when fully adjusted as above, and 1.08-1.13 when also adjusted for non-HDL (p<0.001). HRs were 1.13-1.16 adjusted for LDL, and 1.22-1.26 adjusted for total cholesterol and triglycerides (p<0.001). Splines showed progressively increasing 6-year CHD rates with lower HDL down to 0.5mmol/l. Conclusions This study suggests that lower levels of non-HDL:HDL are a better risk marker for CHD than LDL-cholesterol below 3mmol/l.