We evaluated the relation between early infant diet and insulin-dependent diabetes risk with a meta-analysis of 17 case-control studies. A summary of all studies indicated a moderate effect for exposure to breast-milk substitutes [odds ratio (OR) = 1.38; 95% confidence interval (CI) = 1.18-1.61] and cow's milk-based substitutes (OR = 1.61; 95% CI = 1.31-1.98) before 3 months of age. Fourteen studies relied on retrospectively collected infant diet data based on long-term maternal recall, which may be biased or inaccurate; three studies used existing infant diet records to assess exposure, thus lessening the possibility of recall bias or inaccurate data. The studies using existing records demonstrated little association compared with the studies relying on long-term recall. Studies in which the controls had a participation rate that was more than 20% lower than that of the cases showed a stronger diabetogenic effect of never being breast-fed (OR = 1.58) than studies whose cases and controls had similar participation rates (OR = 1.06). Thus, differences in the participation rates of cases and controls may have biased the results of these studies. This meta-analysis indicates that the weak association between infant diet and risk of diabetes mellitus may have methodologic explanations.