The association between alcohol and blood glucose levels, and whether it is modified by other variables, was examined in a cross-sectional survey of 5518 staff aged 40-65 years at worksites in Auckland and Tokoroa, New Zealand. Diabetes was determined by oral glucose-tolerance tests using 1999 WHO criteria. Usual alcohol intake in the previous 3 months, measured by food frequency questionnaire, was related positively with fasting triglycerides and high-density-lipoprotein (HDL)-cholesterol, and unrelated with fasting glucose, but had an approximate U-shaped relationship with 2-h glucose, which varied from an adjusted mean (S.E.) of 5.62 (0.08) mmol/l in non-drinkers, down to 5.34 (0.08) mmol/l in light alcohol drinkers (alcohol or =20 g per day). Adjusting further for triglycerides increased the mean difference in 2-h glucose for all drinking categories compared with non-drinkers, particularly for heavy drinkers (> or =20 g per day), from -0.22 (S.E. = 0.10) to -0.37 (0.10) mmol/l. The confounding effect of triglycerides suggests alcohol may affect the diabetes risk by a mechanism related to the triglyceride metabolism, which in heavy drinkers may counteract the protective effect of improved insulin sensitivity, resulting in the U-shaped relationship between alcohol and diabetes described in previous studies.