The analysis involved 861 new verified diagnoses of diabetes during 98,543 person-years of follow up (Mean = 13.3 years) among 7802 participants without evidence of a serious chronic disease in the ARIC cohort. The results were robust in that the relative risks were not materially changed in four sensitivity analyses which tested the effects of: 1) excluding participants with evidence of preexisting and preclinical serious chronic disease when their chocolate intake was assessed, as opposed to only excluding participants with evidence of preexisting serious chronic disease; 2) using only chocolate consumption data at exam 1 as predictor, rather than using the data at both exam 1 and 3; 3) including three variables in the causal pathway between chocolate consumption and diabetes risk as extra confounders in the regression model; and 4) adjusting for dietary carbohydrate and fat intake directly instead of by means of an index of dietary quality.