AIMS: To determine whether nuclear magnetic resonance (NMR)-determined lipoprotein profiles predict type 2 diabetes. METHODS: Subjects were 813 male and female participants in the Melbourne Collaborative Cohort Study, aged 40-69 years at baseline (1990-1994), and with a baseline fasting plasma glucose <7.0 mmol/L. Incident type 2 diabetes was identified in 1994-1998 by self-report and confirmation from doctors. Eligible cases and a random group of controls were selected, with NMR data available for 59 cases and 754 non-cases. RESULTS: Concentration of very low density lipoprotein (VLDL) particles (positive) and high density lipoprotein (HDL) particle size (negative) were selected by stepwise regression as predictors of type 2 diabetes. These associations were independent of other non-lipid risk factors, but not plasma triglycerides. Factor analysis identified a factor from NMR variables, explaining 47% of their variation, and characterized by a positive correlation with VLDL, particularly large and medium sized; more low density lipoprotein (LDL) that were smaller; and relatively smaller, but not more HDL particles. This factor was positively associated with diabetes incidence, but not independently of triglycerides. CONCLUSIONS: We identified an atherogenic NMR lipoprotein profile in people who developed diabetes, but this did not improve diabetes prediction beyond conventional triglyceride levels.