The purpose of this study was to determine whether plasma glucose variability, irrespective of glycated hemoglobin (HbA1c), was able to predict the onset of retinopathy in Type 2 diabetic patients. The study was based on a cohort of 130 Type 2 diabetic patients without retinopathy recruited from June 1994 to June 1998. The fundus was reexamined between November 2000 and June 2001, with a mean follow-up period of 5.2 years. Fasting plasma glucose (FPG) variability was measured by its variation coefficient (VC). Stratified and multivariate models were used to estimate the effect of FPG variability and mean HbA1c during follow-up on cumulative incidence (IP) of retinopathy. The IP of retinopathy was 36.2% and increased all along the quartiles of FPG variability (P=.001). In multivariate analyses, only the last quartile of the distribution of VC (OR=3.68; 95% confidence interval (CI) 1.01-13.4; P=.049) was significant. The term of interaction between mean HbA1c and VC was not significant. FPG variability fulfills criteria to be considered a risk factor for retinopathy: A statistically significant association exists after adjustment for confounders, time sequence, dosage response gradient, and biological plausibility.