Objective This study sought to determine the optimal timing of ultrasound in the third trimester to predict birth weight accurately in diabetic women with a singleton pregnancy. Methods A retrospective cohort study of all diabetic women with a singleton pregnancy treated in Halifax, Nova Scotia, was performed. Estimated fetal weight was derived from ultrasound measures using the Hadlock2 equation. The Mongelli equation was used to predict birth weight. The association between gestational age at ultrasound and accuracy of predicted birth weight was assessed, with accuracy as a continuous variable representing the difference between predicted and actual birth weight and as a categorical variable (with four gestational age categories) representing whether predicted birth weight was within, over, or under 250 g of actual birth weight Results The cohort of 943 women comprised 121 (12.8%) with type 1 diabetes, 111 (11.7%) with type 2 diabetes, and 711 (75.4%) with gestational diabetes. Ultrasound scans performed at term were the most accurate in predicting birth weight. At this gestational age, the mean difference between predicted and actual birth weight was −30 g (95% confidence interval −109 to −48). After adjusting for maternal body mass index, age, smoking, type of diabetes, and interval between ultrasound examination and delivery, accuracy improved as gestational age at ultrasound increased (P = 0.005). The odds of underpredicting or overpredicting birth weight were not significantly affected by the timing of the ultrasound examination. Conclusion Because the predictive accuracy of ultrasound prediction of birth weight improves with gestational age, fetal growth assessment at term is recommended to aid with delivery planning in women with diabetes.