This research connects two seemingly unrelated facts that have recently been documented in developing countries, with important consequences for global health: (i) the weak association between nutritional status and income, and (ii) the elevated risk of diabetes among normalweight individuals. The model that we develop to reconcile these facts is based on a set point for body size that is adapted to (low) pre-modern food supply, but subsequently fails to adjust to rapid economic change. During the process of development, some individuals thus remain at their low-BMI set point, despite the increase in their income (food consumption), while others who have escaped their set point (but are not necessarily overweight) are at increased risk of diabetes. The model is tested along different dimensions with multiple data sets. Our analysis indicates that many lean diabetics in developing country populations will be close to their individual-specific set point, suggesting a promising approach to diabetes control (reversal) that involves relatively little weight loss.