OBJECTIVE: Early pregnancy losses increase with marked hyperglycemia in diabetic pregnancy. However, mean loss rates do not differ from those of nondiabetic pregnancy. This observation might be explained by increased fetal losses at the extremes of glycemia in diabetic and nondiabetic pregnancy. To test this hypothesis, we examined relationships of proximate measures of prior glycemia, glycated protein and fructosamine, to pregnancy loss. RESEARCH DESIGN AND METHODS: A total of 389 diabetic and 429 nondiabetic pregnant subjects participated in the Diabetes In Early Pregnancy study. Glycated protein and fructosamine measurements were standardized as multiples of control values for each center (Z score). The logarithm of odds of pregnancy loss were plotted against Z scores and tested by logistic models. RESULTS: Mean pregnancy loss rates were 12% in diabetic and 13% in normal pregnancies. However, over six intervals of glycated protein in diabetic pregnancy, fetal loss rates at the upper and lower extremes (24 and 33%, respectively) were approximately threefold higher than the four intervening rates (8-14%). The odds ratio of pregnancy loss for these extreme intervals to the intervening intervals is 3.0 (P = 0.01). Nondiabetic losses showed a similar pattern. In confirmation, logit pregnancy losses were increased in a J-shaped curve at the glycemic extremes in normal (P < 0.019) and diabetic (P < 0.015) pregnancy. The upper glycemic extreme in diabetic pregnancy was two- to fivefold higher than in control pregnancy. CONCLUSIONS: Pregnancy losses are increased at the extremes of glycemia in both normal and diabetic pregnancy but at higher levels in diabetic pregnancy. The data suggest defensive adaptations against hyperglycemia in diabetic pregnancy.